A three-coordinate radar double-edge feature clustering track condensation method

Abstract

The application provides a three-coordinate radar double-edge feature clustering track condensation method, and belongs to the field of signal processing of three-coordinate radars. Firstly, a two-dimensional plane pool formed by a distance and an elevation of each bearing slice is utilized to find a local maximum value of an amplitude of an echo, distance and elevation cells corresponding to the local maximum value are taken as clustering centers, K-means clustering is performed on the double-edge features of the amplitude distribution and Doppler information of the echo in the distance and the elevation, a one-time track cluster on the bearing slice is obtained, secondary sliding window clustering is performed on the amplitude distribution and Doppler information of the one-time track cluster on adjacent bearing slices, and after the clustering result is obtained, echo data in each class is condensed into a track. The application utilizes two-dimensional characteristic information of the amplitude and the Doppler frequency of the track, suppresses the influence of stray echoes on real targets, makes the echo condensed by the target track purer, and extracts high-precision track position information.

Description

Technical Field

[0001] This invention relates to signal processing technology in the field of three-coordinate radar technology. Background Technology

[0002] Social development and technological innovation have placed higher demands on modern radar. Besides measuring basic information such as target range, azimuth, and elevation, it also needs to detect, acquire, and track targets in complex environments. The processed echo traces contain real target traces and other traces caused by false alarms. The most basic function of trace aggregation technology is to compress the information in the original video echo, distinguish and merge target traces according to trace resolution criteria or methods, filter out false traces, and finally aggregate by amplitude weighting to extract high-precision target trace information.

[0003] The earliest point-of-sight (POS) convergence systems appeared on some advanced radars in the late 1980s and early 1990s. Radars equipped with POS convergence processing systems saw improvements in both point-of-sight accuracy and track performance. Numerous international papers and research findings have addressed POS processing technology in radars with target detection and processing systems. Similar theoretical research emerged in China in the early 1990s. In the 21st century, more and more radars in China have added POS convergence functionality, and engineering improvements to POS convergence methods have continued. The echo penetration method proposed by Xiong Yi et al. reduces the probability of filtering out weak targets during the POS convergence process and effectively reduces the splitting of points in range and azimuth. Ming Liang et al. incorporated multi-dimensional information such as signal-to-noise ratio and track into the POS convergence process, reducing the false alarm rate in automatic target tracking. Wang Xin Guo et al. proposed a Space-Time Crossing Fusion (STCF) method to improve the positional accuracy of radar POS convergence. Research on radar data processing technology in China started later than in other countries, resulting in fewer academic papers and findings on POS convergence processing technology. However, with the development of radar technology and the increasing demands on radar performance, research on POS convergence technology is receiving increasing attention. With the development of electronic technology, various high-performance devices such as DSP and PowerPC are constantly being improved and upgraded. This has led to an increase in the data rate and algorithm complexity of radar information processing systems. Therefore, designing more effective, more universal, and more accurate point aggregation algorithms is of great significance for improving the adaptability and tracking capabilities of modern radar. Summary of the Invention

[0004] To address the problems of point fragmentation and false points caused by traditional point clustering algorithms, this invention proposes a three-coordinate radar bilateral feature clustering point clustering method.

[0005] The present invention proposes a two-sided feature clustering point aggregation method for three-coordinate radar, the technical solution of which includes: Based on the azimuth and elevation wave positions and range resolution of the radar search, the points are divided into azimuth patches and elevation and range cells. The radar echoes are filled into the corresponding point pools in real time. First, K-means clustering is performed using the bilateral features of the amplitude distribution and Doppler information of the echoes in the range and elevation dimensions to obtain primary point clusters on the azimuth patch. Then, secondary sliding window clustering is performed on the amplitude distribution and Doppler information of the primary point clusters on adjacent azimuth patches. After obtaining the clustering results, the echo data in each class is agglomerated into points.

[0006] Furthermore, the bilateral feature K-means clustering method based on amplitude distribution and Doppler information in the distance and elevation dimensions includes: dividing the points on a single azimuth slice into cells in the distance and elevation dimensions; finding K points with the largest amplitudes as cluster centers using a nine-square grid labeling method; calculating the Euclidean distance and Doppler frequency difference from each of the remaining points to the K center points; selecting the cluster center point with the closest distance and a Doppler frequency difference that meets a certain threshold as the label category; recalculating the cluster center; iterating multiple times until the Euclidean distance between the new center point and the original center point meets the threshold, thereby completing the classification and merging of the points on the azimuth slice.

[0007] Furthermore, the secondary sliding window clustering method includes: for each point cluster in the current azimuth slice, based on the index number of the elevation angle grid and distance grid of its cluster center, searching for the cluster center point in the corresponding 3x3 grid in the next azimuth slice. If two conditions are met simultaneously: 1) the Doppler frequency difference meets a certain threshold; 2) the amplitude envelope formed with the points in the current point cluster has a certain continuity and basically conforms to the Gaussian curve, then the cluster center point is added to this point cluster; if no point is found or the azimuth expansion of the point cluster exceeds the set upper limit, then the centroid method is used to calculate the cluster center of the point cluster and output it for reporting, while adding the unassociated cluster center points in the next azimuth slice to the cluster center cluster; repeating the above steps to complete the point classification and merging between azimuth slices.

[0008] This invention utilizes the amplitude and Doppler frequency of the target trace to suppress the influence of clutter echoes on the real target, resulting in a purer echo from the target trace and thus extracting high-precision trace position information. This method reduces the impact of clutter echoes on target detection, decreases the occurrence of trace splitting and the formation of false traces, ultimately reducing the formation of false tracks at the back end and improving the stability of track tracking. The process is clear and simple to implement, and it has significant engineering application value for signal processing in three-coordinate radar. Attached Figure Description

[0009] Appendix Figure 1 Flowchart of the implementation of this invention. Detailed Implementation

[0010] The present invention will now be explained and described in conjunction with the accompanying drawings and embodiments.

[0011] Modern three-coordinate radars employ technologies such as moving target detection, adaptive moving target display, and pulse Doppler signal processing. The processed point traces typically contain not only conventional azimuth, range, elevation, and amplitude information, but also Doppler frequency information. To fully utilize the amplitude and Doppler frequency characteristics of the traces, suppress the influence of cluttered echoes on the true target, and extract high-precision point trace location information, this invention divides the point traces into azimuth patches and elevation / range cells based on the radar's search azimuth and elevation wave positions and range resolution. The radar echoes received are filled into the corresponding point trace pools in real time. First, K-means clustering is performed using the bilateral features of the echo amplitude distribution and Doppler information in the range and elevation dimensions to obtain primary point trace clusters on the azimuth patch. Then, secondary sliding window clustering is performed on the amplitude distribution and Doppler information of primary point trace clusters on adjacent azimuth patches. After obtaining the clustering results, the echo data within each cluster is aggregated into the final point trace.

[0012] The preferred implementation process of the three-coordinate radar bilateral feature clustering point aggregation method proposed in this invention is as follows: Figure 1 As shown, the specific process is described as follows: The point data packets obtained by a three-coordinate radar after undergoing a series of signal processing measures such as anti-asynchronous processing, pulse compression, MTI / MTD are processed by point clustering. Each point contains unique amplitude, range, elevation angle and Doppler frequency information.

[0013] S1. Divide the azimuth slices according to the radar's preset scanning azimuth wave position, and sequentially retrieve the point data packets from the point data pool and store them into the corresponding azimuth slice buffer.

[0014] S2. After collecting the beam data of a azimuth slice, take the spot data of that azimuth slice and divide it into cells according to elevation angle and range. The size of the elevation angle cell depends on the beam width at the elevation angle, and the size of the range cell is related to the range resolution of the radar. Each spot falls in the corresponding cell, and the spots in the same cell are grouped and merged according to Doppler frequency.

[0015] S3. Select K points from the points in the first orientation patch. The specific steps for using the initial cluster centers are as follows.

[0016] 1) Traverse all points and select the point with the largest amplitude as... P 1. Mark the 3x3 grid containing that point as invalid.

[0017] 2) Continue to select the point with the largest amplitude from the remaining valid cells and add it to the cluster center set ψ. Similarly, mark the cell containing that point as invalid.

[0018] 3) Repeat step 2) until K points have been selected or all cells have been invalidated.

[0019] S4. Calculate the Euclidean distance and Doppler frequency difference from each of the remaining points to the center point of K, and select the point with the shortest distance and a Doppler frequency difference that meets a certain threshold. f limit One of the cluster centers is used as the label category, i.e. .

[0020] S5. Recalculate the new centroid of each cluster using the centroid method.

[0021] S6. If the calculated Euclidean distances between the new center point and the original center point are both less than a certain threshold... d limit If the condition is met, then execute step 7; otherwise, repeat steps S4-S6.

[0022] S7. After completing the classification and merging of the points on the azimuth map, record the index number of the elevation angle grid and the distance grid of the cluster center point of each point cluster.

[0023] S8. Using the same method, cluster the points of the next azimuth patch in the distance and elevation dimensions to obtain the cluster center clusters. .

[0024] S9, targeting ψ current For each point cluster in the array, based on the index number of the elevation grid and distance grid of its cluster center, in the next azimuth slice ψ next Find the cluster center point within the corresponding 3x3 grid. If two conditions are met simultaneously: 1) the Doppler frequency difference meets a certain threshold. d limit 2) If the amplitude envelope formed by the points in the current point cluster has a certain continuity and basically conforms to the Gaussian curve, then the cluster center point is added to this point cluster and the cluster center is recalculated.

[0025] S10. If no point is found or the azimuth expansion of the point cluster exceeds the set upper limit, calculate the cluster center of the point cluster using the centroid method and output it for reporting, and remove the point cluster from ψ current Delete it.

[0026] S11, for ψ next Unassociated cluster centers P Add it to the cluster center cluster ψ current .

[0027] S12. Repeat S8 to S11 until all orientation slices have been traversed, then... current All cluster centers are reported.

Claims

1. A method for clustering point traces using two-sided features in a three-coordinate radar, characterized in that: Based on the azimuth and elevation wave positions and range resolution obtained by the radar search, the points are divided into azimuth patches and elevation / range cell blocks. The radar echoes are filled into the corresponding point pools in real time. First, K-means clustering is performed using the bilateral features of the echo amplitude distribution and Doppler information in the range and elevation dimensions to obtain primary point clusters on the azimuth patch. Then, secondary sliding window clustering is performed on the amplitude distribution and Doppler information of the primary point clusters on adjacent azimuth patches. After obtaining the clustering results, the echo data within each cluster is aggregating into points. The secondary sliding window clustering includes: for each point cluster in the current azimuth patch, based on the elevation angle of its cluster center... Using the index of the grid and the distance grid, the cluster center point is searched within the corresponding 3x3 grid in the next azimuth slice. If two conditions are met simultaneously: 1) the Doppler frequency difference meets a certain threshold; 2) the amplitude envelope formed with the points in the current trace cluster has a certain continuity and basically conforms to the Gaussian curve, then the cluster center point is added to this trace cluster. If no point is found or the azimuth expansion of the trace cluster exceeds the set upper limit, the centroid method is used to calculate the cluster center of the trace cluster and output it for reporting. At the same time, the cluster center points in the next azimuth slice that are not associated are added to the cluster center cluster. Repeat the above steps to complete the trace classification and merging between azimuth slices.

2. The method for clustering point traces using two-sided features of a three-coordinate radar according to claim 1, characterized in that: The K-means clustering includes: dividing the points on a single azimuth slice into cells along the distance and elevation dimensions; finding the K points with the largest amplitudes using a nine-square grid method as cluster centers; calculating the Euclidean distance and Doppler frequency difference from each of the remaining points to the K center points; selecting the cluster center point with the closest distance and a Doppler frequency difference that meets a certain threshold as the labeled category; recalculating the cluster centers; iterating multiple times until the Euclidean distance between the new center point and the original center point meets the threshold, thereby completing the classification and merging of the points on the azimuth slice.

Images