A method for UHF radar river vessel extended target clustering

Abstract

The application discloses a method for UHF radar river ship extended target clustering. The method utilizes the idea of three-dimensional grid division, firstly normalizes three-dimensional information according to the resolution of distance, Doppler and azimuth angle, and constructs a three-dimensional grid; then, pre-clusters by expanding the resolution of each dimension to remove obvious discrete interference points; finally, re-clusters by utilizing the relationship between the motion state of the river ship and the gradient descent direction of the constructed three-dimensional grid to obtain the final river ship extended target clustering center. The method can improve the clustering accuracy, is effectively applicable to various scenes, and can make the UHF radar river ship detection system have higher universality and stability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar, and particularly relates to a method for UHF radar river ship extended target clustering. BACKGROUND

[0002] In recent years, ultrahigh frequency (UHF) radar has occupied a very important position in river hydrological monitoring, and can be used to measure river surface flow velocity, river flow and other parameters, and has become a relatively mature technology in river monitoring. With the development of society, tourism and transportation cannot do without the ship as a carrier in the inland river, so the detection and positioning of the ship target in the river are of great significance for the related departments to schedule, warn and dredge the river ship. In this case, the UHF radar has the characteristics of "one machine with multiple uses" that can monitor and position the ship target in the river while monitoring the river hydrology, so it has become a very effective method for detecting the ship target in the river.

[0003] At present, there are many methods for detecting the river ship, such as synthetic aperture radar pattern, satellite optical image, video monitoring, wireless sensor network, etc., but they have some limitations more or less. The synthetic aperture radar system is large in cost and complex in erection and use; the satellite optical image and video monitoring are very dependent on the external environment, and may affect the monitoring efficiency when encountering bad conditions such as heavy fog or rain; the wireless sensor network has a small detection range, and needs to erect multiple sensors to connect into a network, which has a high cost. The UHF radar has become a very suitable method for detecting the river ship target due to its low transmission power, large detection range, small and easy-to-erect equipment and convenient use.

[0004] For the UHF radar river ship target detection, the ship target occupies multiple resolution units due to the high resolution of the UHF radar, thereby forming an extended target. The improved constant false alarm rate method can detect these extended target detection points, but cannot obtain specific ship target parameters such as distance and radial velocity. Therefore, it is still necessary to cluster the ship extended target detection points, consider the clustering center as the target center of the ship target, and thereby obtain the corresponding ship target parameters.

[0005] There are many methods for the detection of extended targets at present, but because of the problems of needing multi-dimensional information clustering, the uniformity of detection points density, and the inability to know the number of categories in advance, the clustering of extended target detection points of ultra-high frequency radar exists, so most methods are not suitable for the scene of ultra-high frequency radar extended target detection. A few applicable clustering methods are mainly divided into two-dimensional and three-dimensional clustering methods, two-dimensional is the distance and radial velocity of the target, and three-dimensional is the addition of the target azimuth. Because the two-dimensional clustering method does not use the azimuth information of the target, it cannot distinguish multiple targets in the target scene where the distance and radial velocity are similar and the azimuth is different. Although the three-dimensional clustering method uses the azimuth information of the target, it only simply integrates the three-dimensional information of the target, and cannot distinguish multiple targets in the case where the distance, radial velocity and azimuth of the target are similar. Therefore, developing an ultra-high frequency radar extended target clustering method that can adapt to more scenes has become a key problem of the ultra-high frequency radar river ship target detection. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a method for UHF radar river ship extended target clustering, which can cope with more scenes and improve the stability and universality of UHF radar river ship target detection.

[0007] In order to achieve the above purpose, the technical scheme provided by the present application is a method for UHF radar river ship extended target clustering, comprising the following steps:

[0008] Step 1, monitoring the river ship by UHF radar to obtain the ship target echo data, obtaining the distance-Doppler R-D diagram of the target echo and the distance and radial velocity information of the target by distance resolution and velocity resolution, and obtaining the extended target detection points by a constant false alarm rate detector;

[0009] Step 2, obtaining the target azimuth information by an azimuth estimation algorithm, combining the target distance and radial velocity information obtained in step 1 to obtain the distance, radial velocity and azimuth information of each extended target detection point in step 1;

[0010] Step 3, constructing a three-dimensional grid containing all the extended target detection points obtained in step 1;

[0011] Step 4, dividing the three-dimensional grid obtained in step 3 according to a certain resolution to obtain a plurality of grid blocks, and calculating the self attribute parameters of each grid block;

[0012] Step 5, removing the discrete grid blocks in step 4, and pre-clustering the grid blocks that meet the clustering conditions to obtain a pre-clustering result;

[0013] Step 6, re-clustering the grid blocks after pre-clustering in step 5, and taking the distance, radial velocity and azimuth information of each cluster center after re-clustering as the corresponding information of the ship extended target;

[0014] Step 6.1, in the cluster set ∑ obtained by pre-clustering, for each cluster β containing several three-dimensional grid blocks in the set ∑ i , scanning from top to bottom in the azimuth dimension;

[0015] Step 6.2, scanning the cluster β i in step 6.1, and performing association judgment on the last associated detection point c in each cluster γ i in the newly generated cluster set Ω corresponding to the to-be-detected point b in the azimuth a in the cluster β

[0016] Step 6.3, after each detection point in the cluster β i is associated, finding the next unassociated cluster β j , performing the association process of step 6.2 on the cluster β j , and ending the clustering process until all the detection points in the clusters β in the set ∑ are associated.

[0017] Moreover, the three dimensions of the three-dimensional grid in step 3 are distance dimension, Doppler dimension and azimuth dimension, respectively, the Doppler dimension corresponds to radial velocity information, and the coordinates T of the extended target point in the three-dimensional grid are obtained by the following formula:

[0018] T = {(m, n, θ) | m ∈ [1, M], n ∈ [1, N], θ ∈ [-90°, 90°]} (1)

[0019]

[0020] In the formula, M and N are the total number of Doppler dimension and distance dimension units respectively, K is the total number of extended target detection points, R, V and A represent the distance, radial velocity and azimuth of the target respectively, and the corresponding resolutions are Δr, Δv and Δa respectively. The azimuth resolution Δa is set to 1°, and the distance resolution and radial velocity resolution are respectively:

[0021]

[0022]

[0023] In the formula, Δτ is the time delay resolution, c is the electromagnetic wave propagation speed, B is the effective bandwidth, Δf d is the Doppler resolution, f0 is the radar center frequency, λ is the radar signal wavelength, and τ' is the pulse width.

[0024] Moreover, the self attribute parameters of each grid block in step 4 include: the number of extended target detection points contained in the grid block, the distance dimension mean and range of the extended target detection points, the Doppler dimension mean and range of the extended target detection points, and the azimuth angle dimension mean and range of the extended target points.

[0025] Moreover, the condition of the discrete grid block in step 5 is that the number of extended target detection points is less than N1, and the range of one dimension among the three dimensions is greater than N2 units; the condition of the associated grid block is that the distance dimension mean difference between two grid blocks is less than N3 units, the Doppler dimension mean difference is less than N4 units, and the azimuth angle dimension distance mean is less than N5 units; the discrete grid block is removed, and the grid blocks meeting the distance, radial velocity and azimuth angle threshold values are associated to obtain a pre-clustering result.

[0026] Moreover, the association judgment in step 6.2 includes the following four judgment conditions: 1) the difference of the Doppler dimension between b and c in the three-dimensional grid is within the Doppler dimension threshold T m ; 2) the difference of the distance dimension between b and c in the three-dimensional grid is within the distance dimension threshold T n ; 3) the difference of the azimuth angle dimension between b and c in the three-dimensional grid is within the azimuth angle dimension threshold T θ ; and 4) the direction of c pointing to b in the three-dimensional grid is the same as the gradient descent direction of c in the cluster γ i ; if the four judgment conditions are all met, b is associated with c, and b becomes the last associated detection point in the cluster γ i ; if each cluster γ i in Ω cannot be associated with b, a new cluster γ i ' is generated with b as the starting point and is added into Ω.

[0027] Compared with the prior art, the present application has the following advantages:

[0028] The three-dimensional information of the distance, radial velocity and azimuth angle of the target can be fully utilized, the target motion state is associated with the parameter change trend in the target extended range, different targets can be distinguished when the above three-dimensional information is close, and the generality of the UHF radar river ship detection can be effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a flowchart of an embodiment of the present application.

[0030] Figure 2 is a schematic diagram of a river ship extended target detection point provided by an embodiment of the present application.

[0031] Figure 3 is a schematic diagram of a three-dimensional grid provided by an embodiment of the present application.

[0032] Figure 4is a schematic diagram of a clustering algorithm process provided by an embodiment of the present application, wherein Figure 4 (a) represents dividing a three-dimensional grid into grid blocks, Figure 4 (b) represents an initial clustering result, Figure 4 (c) represents re-clustering the initial clustering result, Figure 4 (d) represents a re-clustering result.

[0033] Figure 5 is a river ship target motion state diagram provided by an embodiment of the present application.

[0034] Figure 6 is a comparison diagram of a target actual position in an R-D diagram and results obtained by the method of the present application and two other clustering methods, wherein Figure 6 (a) represents an actual distribution of the target, Figure 6 (b) represents a clustering result obtained by the method of the present application, Figure 6 (c) represents a clustering result obtained by a three-dimensional clustering method, Figure 6 (d) represents a result obtained by a grid-based clustering method. DETAILED DESCRIPTION

[0035] The present application provides a method for UHF radar river ship extended target clustering, and the technical solutions of the present application are further described below with reference to the drawings and embodiments.

[0036] As shown in Figure 1 , the present application provides a method for UHF radar river ship extended target clustering, comprising the following steps:

[0037] Step 1: Monitor the river ship by UHF radar to obtain ship target echo data, obtain the Range-Doppler (R-D) of the target echo and the distance and radial velocity information of the target by range resolution and velocity resolution, and obtain the extended target detection point by a constant false alarm rate detector.

[0038] The UHF radar transmitter generates a transmission signal, which is transmitted by a transmitting antenna. When the transmission signal encounters a ship target, it is reflected and scattered to generate a target echo. The target echo is received by a receiving antenna and transmitted to a receiver for preprocessing, and then transmitted to a host computer for processing. Thereafter, signal processing is performed on the host computer. The target echo signal is first subjected to range resolution and velocity resolution to obtain an R-D diagram, and the distance and radial velocity information of the target is obtained. Then, constant false alarm rate (CFAR) detection is performed in the R-D diagram to obtain the detection point of the extended target, as shown in Figure 2 . Figure 2The left side of the middle represents a group of target detection points on the left side, and the right side of the middle represents a group of target detection points on the right side. The group of target detection points on the left side includes detection points of target 1 and target 4, and the group of target detection points on the right side includes detection points of target 2 and target 3. It can be seen that it is difficult to distinguish targets 1 and 4 or targets 2 and 3 in the R-D diagram, which have close radial velocities and distances.

[0039] Step 2: Obtain target azimuth information by an azimuth estimation algorithm, and combine the target distance and radial velocity information obtained in step 1 to obtain distance, radial velocity and azimuth information of each extended target detection point in step 1.

[0040] For each extended target detection point, a multiple signal classification (MUSIC) algorithm is used to obtain its azimuth information.

[0041] Step 3: Construct a three-dimensional grid including all the extended target detection points obtained in step 1.

[0042] The constructed three-dimensional grid is shown in FIG. 2. Figure 3 The three dimensions are distance dimension, Doppler dimension and azimuth dimension, wherein the Doppler dimension corresponds to the radial velocity information. Figure 3 The left side of the middle represents a three-dimensional display of the group of target detection points on the left side, and the right side of the middle represents a three-dimensional display of the group of target detection points on the right side. Figure 2 The left side of the middle represents a three-dimensional display of the group of target detection points on the left side, and the right side of the middle represents a three-dimensional display of the group of target detection points on the right side. Figure 2 The left side of the middle represents a three-dimensional display of the group of target detection points on the left side, and the right side of the middle represents a three-dimensional display of the group of target detection points on the right side.

[0043] T = {(m, n, θ) | m ∈ [1, M], n ∈ [1, N], θ ∈ [-90°, 90°]} (1)

[0044]

[0045] In the formula, M and N are respectively the total number of Doppler dimension units and distance dimension units, K is the total number of extended target detection points, R, V and A respectively represent the distance, radial velocity and azimuth of the target, and their corresponding resolutions are Δr, Δv and Δa respectively.

[0046] The azimuth resolution Δa is set to 1°, and the distance resolution and the radial velocity resolution are respectively:

[0047]

[0048]

[0049] In the formula, Δτ is the time delay resolution, c is the electromagnetic wave propagation speed, which is approximately the speed of light, B is the effective bandwidth, Δf is the frequency resolution, and c is the electromagnetic wave propagation speed. dHere, f0 is the Doppler resolution, λ is the radar center frequency, τ′ is the radar signal wavelength, and τ′ is the pulse width.

[0050] Step 4: Divide the 3D mesh obtained in Step 3 into several mesh blocks according to a certain resolution, and calculate the property parameters of each mesh block.

[0051] The 3D mesh obtained in step 3 is divided into several mesh blocks, such as Figure 4 As shown in (a). In this embodiment, the distance and Doppler dimensions of the grid blocks are both 8 units long, and the azimuth dimension is 20 units long. The self-attribute parameters of each grid block are calculated, including: the number of extended target detection points contained in the grid block, and the mean and range of the distance dimension of the extended target detection points, as well as the mean and range of the Doppler dimension and the azimuth dimension of the extended target points.

[0052] Step 5: Remove the discrete grid blocks from Step 4, and perform pre-clustering on the grid blocks that meet the clustering conditions to obtain the pre-clustering results.

[0053] Because the extended target detection points are numerous and concentrated, while the interference detection points are few and scattered, this characteristic can be used to remove discrete grid blocks that clearly only contain interference points. Furthermore, grid blocks that meet the distance, radial velocity, and azimuth thresholds are associated to obtain pre-clustering results. In this embodiment, the conditions for discrete grid blocks are: the number of extended target detection points is less than 5, and the range of one of the three dimensions is greater than 10 units; the conditions for associated grid blocks are: the mean difference in distance dimension is less than 6 units, the mean difference in Doppler dimension is less than 8 units, and the mean distance in the azimuth dimension is less than 20 units. The pre-clustering results are as follows: Figure 4 As shown in (b).

[0054] Step 6: Re-cluster the mesh blocks that were pre-clustered in Step 5, and use the distance, radial velocity, and azimuth information of each cluster center after re-clustering as the corresponding information of the ship's extended target.

[0055] like Figure 4 As shown in (c), the pre-clustering results obtained in step 5 are re-clustered to separate the results that may contain multiple clusters in the pre-clustering results, resulting in the re-clustering results as shown in the figure. Figure 4 As shown in (d).

[0056] Generally, if we assume that the river vessel is moving at a constant velocity in a straight line along the river channel, then the following holds true:

[0057] v r =v sinθ (5)

[0058] In the formula, v ris the actual velocity of the ship, θ is the target azimuth angle, i.e. the angle between the line connecting the radar and the target and the normal direction of the radar, and θ ∈ (-90°, 90°), and sinθ is monotonically increasing.

[0059] The change of the radial velocity with the azimuth angle only depends on the direction of the actual velocity. If the actual velocity is in the positive direction, the radial velocity decreases with the azimuth angle. If the actual velocity is in the negative direction, the radial velocity increases with the azimuth angle. Specifically, Figure 5 v r1 ≈v r2 , R1≈R2, at this time, the ship1 and ship2 extended target points are close in the R-D diagram, and it is difficult to distinguish them. However, it can be known from Figure 5 that the two ships are actually far apart and are easy to distinguish, because θ1 is negative and θ2 is positive at this time. If the two ships continue to move and meet at the position of the normal line of the radar, at this time θ1≈θ2, it is not possible to directly determine by θ. Since the change of the radial velocity with the azimuth angle only depends on the direction of the actual velocity, taking the direction of ship2 as the positive direction and the direction of ship1 as the negative direction, the radial velocity of the ship1 extended target point will increase with the azimuth angle, and the radial velocity of the ship2 extended target point will decrease with the azimuth angle. Therefore, in the intersection scenario, two ship targets can be distinguished according to the change trend of the radial velocity of the target with the decrease of the azimuth angle. In the grid block after removing the discrete grid block, the azimuth angle dimension is scanned from top to bottom, and the detection points belonging to different extended targets can be distinguished by detecting the change of the point Doppler velocity unit with the decrease of the azimuth angle.

[0060] Step 6.1, in the cluster set ∑ obtained by pre-clustering, for each cluster β i (β i ∈∑) containing several three-dimensional grid blocks, the azimuth angle dimension is scanned from top to bottom.

[0061] Step 6.2, the to-be-detected point b corresponding to the azimuth angle a in the cluster β i scanned in step 6.1 is associated with the last associated detection point c in each cluster γ i (γ i ∈Ω) in the newly generated cluster set Ω of re-clustering.

[0062] The association judgment includes the following four judgment conditions: 1) the difference between b and c in the Doppler velocity dimension in the three-dimensional grid is within the Doppler velocity threshold T m , 2) the difference between b and c in the distance dimension in the three-dimensional grid is within the distance threshold T n , 3) the difference between b and c in the azimuth angle dimension in the three-dimensional grid is within the azimuth angle threshold T θIn, 4). The direction of c pointing to b in the three-dimensional grid and the cluster γ that c belongs to i The gradient descent direction is the same. If the four judgment conditions are all satisfied, b is associated with c, and b becomes the cluster γ i The last associated detection point in γ. If every cluster γ in Ω i cannot be associated with b, a new cluster γ i containing b is generated and added to Ω. In this embodiment, T m , T n , and T θ are set to 4 units, 2 units, and 7 units, respectively.

[0063] Step 6.3, after every detection point in the cluster β i is associated, the next unassociated cluster β j is found. j The association process of step 6.2 is performed on the cluster β

[0064] The target actual distribution of the comparative experiment is shown in Figure 6 (a), and the clustering results obtained by using the grid-based gradient descent (GBGD) method proposed in the present application are shown in Figure 6 (b), Figure 6 (c), and Figure 6 (d) are the results obtained by three-dimensional (3D) clustering and grid-based (GB) clustering methods. From Figure 6 (a)-(d), it can be seen that for the scenes corresponding to targets 2 and 3, other three-dimensional clustering methods cannot distinguish them, while the method proposed in the present application can distinguish them. The offset of the clustering center of each target relative to the actual target center is shown in Table 1:

[0065] Table 1: Comparison of clustering center offset

[0066]

[0067] From Table 1, it can also be seen that for targets 2 and 3, the clustering results obtained by the method proposed in the present application are significantly better than those of the other two three-dimensional clustering methods. In combination with Figure 6 , it can be shown that the method proposed in the present application is effective and versatile, and can make the UHF radar river vessel detection system applicable to more scenes and effectively improve the stability.

[0068] The specific embodiments described herein are presented only to provide illustrations of the aspects of the application. Alternatives to the embodiments described herein can be employed without departing from the spirit or scope of the application. Accordingly, the specifics of the described embodiments are not intended to limit the scope of the application but are presented as being illustrative only.

Claims

1. A method for UHF radar river vessel extension target clustering, characterized by, The method comprises the following steps: Step 1, monitoring river vessels by UHF radar, obtaining vessel target echo data, obtaining the range-Doppler R-D diagram of the target echo and the range and radial velocity information of the target by range resolution and velocity resolution, and obtaining the extended target detection points by a constant false alarm rate detector; Step 2, obtaining the azimuth angle information of the target by an azimuth angle estimation algorithm, combining the range and radial velocity information obtained in step 1 to obtain the range, radial velocity and azimuth angle information of each extended target detection point in step 1; Step 3, constructing a three-dimensional grid containing all the extended target detection points obtained in step 1; Step 4, dividing the three-dimensional grid obtained in step 3 according to a certain resolution to obtain a plurality of grid blocks, and calculating the self attribute parameters of each grid block; Step 5, removing the discrete grid blocks in step 4, and pre-clustering the grid blocks meeting the clustering condition to obtain a pre-clustering result; Step 6, re-clustering the grid blocks after pre-clustering in step 5, and taking the range, radial velocity and azimuth angle information of each clustering center after re-clustering as the corresponding information of the extended target of the vessel; Step 6.1, in the pre-clustering of the set of clusters , each of the clusters , contains several three-dimensional grid blocks , the scanning is performed from top to bottom in the azimuthal angle Step 6.

2. Scan the clusters in step 6.1 the middle azimuth the corresponding detection point to be detected the newly generated cluster set each cluster in the newly generated cluster set the last associated detection point in the cluster make an association judgment; Association judgment includes the following 4 judgment conditions: 1) and The Doppler difference in a 3D mesh is at the Doppler threshold. Within; 2) and In a 3D mesh, the difference in distance dimension is at the distance dimension threshold. Within; 3) and The difference in the azimuth dimension of the three-dimensional mesh is at the azimuth dimension threshold. Within; 4) In the three-dimensional mesh point to direction and Cluster The gradient descent directions are the same; if all four conditions are met, then... and Related, Become a cluster The last associated detection point in the middle; if Each cluster Unable to be with If associated, a result will be generated. New clusters at the beginning join in middle; Step 6.3, cluster After each detection point is associated, find the next unassociated cluster , cluster Perform the association process of Step 6.2 until all clusters are associated, end the clustering process.

2. A method for UHF radar river vessel extension target clustering as claimed in claim 1 characterized by: The three dimensions of the three-dimensional grid in step 3 are distance dimension, Doppler dimension, and azimuth dimension respectively, wherein the Doppler dimension corresponds to radial velocity information, and the coordinates of the extended target point in the three-dimensional grid are are obtained from the following formula: （1） （2） wherein, and are the total number of Doppler and range cells, respectively, is the total number of extended target detection points, denote the range, radial velocity and azimuth of the target, respectively, with corresponding resolutions of the azimuth resolution is set to 1°, and the range and radial velocity resolutions are:​​​​ （3） （4） wherein is the time delay resolution, is the electromagnetic wave propagation speed, is the effective bandwidth, is the Doppler resolution, is the radar center frequency, is the radar signal wavelength, is the pulse width.

3. A method for UHF radar river vessel extension target clustering as recited in claim 1, characterized by: The self attribute parameters of each grid block in step 4 include: the number of extended target detection points contained in the grid block, the range dimension mean and range of the extended target detection points, the Doppler dimension mean and range of the extended target detection points, and the azimuth angle dimension mean and range of the extended target points.

4. A method for UHF radar river vessel extension target clustering as recited in claim 1, characterized by: The condition of the discrete grid block in step 5 is that the number of the extended target detection points is less than , and the range of a certain dimension among the three dimensions is greater than The condition of the associated grid block is that the distance dimension mean difference between the two grid blocks is less than , the Doppler mean difference is less than , and the azimuth angle dimension distance mean is less than The discrete grid block is removed, and the grid blocks satisfying the distance, radial velocity and azimuth angle threshold values are associated to obtain a pre-clustering result.

Images