A sidelobe clutter tracking method based on improved K-means clustering

Abstract

The present application provides a sidelobe clutter tracking method based on improved K-means clustering, which belongs to the field of radar information processing technology. Specifically, it includes a post-processing operation based on a suspected clutter target data set formed after setting an appropriate background level detection, and a clustering method based on a data mining technology according to spatial relationships. For a specific clutter target data set, the classification number K is initialized according to the number of clutter targets in the spectrum. in After iterative calculation of the minimum distance between the sample objects and the cluster center in the target data set, the targets that meet the threshold are classified into one category. After the iteration, the final cluster center distance is less than D. L All classes of , determine the clustering results, and the number of targets in the class is greater than D G And the intra-class distance is less than D R The class is retained as the clutter class. Through the processing solution of the present application, the range and edge of strong sidelobe clutter can be accurately distinguished.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radar information processing, in particular to a sidelobe clutter tracking method based on improved K-means clustering. BACKGROUND

[0002] When a modern airborne pulsed Doppler (PD) radar detects targets in the down view, it mainly uses the medium PRF (MPRF) waveform mode. This mode adopts full coherent PD processing, that is, it can realize detection and tracking of head-on / tail-following targets, and can also complete target monitoring in the full air space of the up view / down view, and is therefore widely used.

[0003] Target detection in this mode is performed in the one-dimensional "range" and two-dimensional "range-frequency" planes. Two-dimensional detection mainly uses the Doppler frequency difference between targets and main clutters to realize frequency separation, so that constant false alarm technology can be used to complete target detection in the clear area, thereby maximizing the detection probability.

[0004] The conventional constant false alarm detection method uses a sliding window to calculate the background power around the detection point, such as the cell average constant false alarm (CA-CFAR) detection processing method, which compares the size relationship between the detection point and the background power to determine the target. In the case of independent and identically distributed Rayleigh envelopes and no interference target echoes or clutter edge in the reference sliding window, this method can obtain quasi-optimal detection performance.

[0005] When the detection cell is located at the edge of the main clutter, the height line clutter, or the strong sidelobe clutter, the background power in the area where the detection cell is located and the strong clutter or interference target echoes mixed in the reference window sliding window will cause changes in the detection threshold. If only the conventional constant false alarm technology is used, it will cause different degrees of detection probability decline, target shadowing, or false alarm probability rise, thereby affecting the normal use of the radar. SUMMARY

[0006] Therefore, the present application provides a sidelobe clutter tracking method based on improved K-means clustering, which solves the problems in the prior art and can accurately distinguish the range and edge of strong sidelobe clutter.

[0007] The sidelobe clutter tracking method based on improved K-means clustering provided by the present application adopts the following technical solution:

[0008] A sidelobe clutter tracking method based on improved K-means clustering, characterized in that it comprises:

[0009] Step 1: divide the RF plane into L N×M rectangular windows, and take the maximum background power position P pos (i) = [nr i ,nf i], i = 1, 2, 3..., L as the subsequent clustering initialization value;

[0010] Step 2, for each detection point on the RF plane, a sliding window processing is performed on the distance direction, and the sliding window width is W, to obtain the background P of each detection point back [nr, nf], nr = 1, 2, 3..., Nr, nf = 1, 2, 3..., Nf;

[0011] Step 3, it is judged whether each signal point is a suspected clutter target point, if the signal point is a suspected clutter target point, the clutter flag of the signal point position is set to 1, the clutter target point Z = Z + 1, otherwise the clutter flag of the signal point position is set to 0, and finally a clutter binary spectrum matrix Clu[nr, nf] is formed and the position information S of all suspected clutter target points is stored pos (i) = [nr i , nf i ], i = 1, 2, 3..., Z;

[0012] Step 4, the number of clutter flags of the signal point positions in each rectangular block which is 1 is counted, if the number meets a preset threshold, it is indicated that the rectangular block is a clutter fast, and the position P of the maximum background power in the rectangular window is recorded pos denoted as Clu pos , the number of rectangular windows meeting the preset threshold is recorded;

[0013] Step 5, the input of the K-means clustering algorithm is set, S pos as the clutter target data set, wherein Z is the size of the clutter data set, the initial number of categories K in = c block , and the iteration number t of the algorithm;

[0014] Step 6, the sample objects in Clu pos are taken as the initial clustering centers of K in categories;

[0015] Step 7, for each sample object S pos (i) in the clutter data set, the clustering center Clu pos (i) with the smallest distance is found, if the smallest distance is smaller than a threshold value D L , the sample object S pos (i) is assigned to the category C pos corresponding to the clustering center Clu i (i), otherwise the sample object is removed from the clutter data set;

[0016] Step 8, the mean Clu pos (i) of the sample objects in each category is calculated as the new clustering center point;

[0017] Step 9, calculate the clustering criterion function;

[0018] Step 10, repeat steps 7, 8, 9 until the specified iteration number t is met or the clustering criterion function converges.

[0019] Optionally, the step 3 includes averaging the background power of all sampling points in each rectangular window

[0020] ,

[0021] Select the minimum background average power T in the L rectangular window average background power back = min(P aver ) as the noise threshold for judging the suspected clutter target, initialize the clutter target point number Z = 0, multiply T back by a preset threshold coefficient K, and then compare it with the background P back [nr, nf] of each detection point, if P back [nr, nf] > K × T back , it is considered that the point is a suspected clutter target point.

[0022] Optionally, the step 4 includes dividing the clutter flag binary spectrum plane into L uniform rectangular windows in the manner in step 1 and initializing the suspected clutter block number c block = 0, summing the flag in each rectangular window and comparing it with a preset threshold K flag , if

[0023] ,

[0024] Record the position of the maximum background power in the rectangular block Clu pos (c block ) = P pos (i), c block = c block + 1.

[0025] Optionally, it further includes step 11, for the completed cluster C = [1, K in ], calculate the distance of all final cluster center points, if less than D L , merge the clusters between them to form a new clutter cluster set;

[0026] Step 12, calculate the distance between the target points in the single cluster in the new cluster set, the target number in each cluster, according to the above results, the clustering result of each cluster G > D G and the distance between the target points in the single cluster R < D R is taken as the clutter tracking target result data set, D G is the target number threshold value, DR The in-class distance threshold value.

[0027] Optionally, the preset threshold K flag Half of the number of target points in the rectangular frame.

[0028] In summary, the present application includes the following beneficial technical effects:

[0029] For a specific clutter target data set, the number of block clutter targets on the spectrum is used to initialize the classification number K in After iterative calculation of the minimum distance between sample objects and cluster centers in the target data set, the targets meeting the threshold are classified into a class, and after the iteration is completed, all classes with a final cluster center distance less than D L are merged to determine the clustering result, and the class with a number of targets greater than D G and an intra-class distance less than D R is reserved as a clutter class.

[0030] In the detection process of the radar air mode, the distribution characteristics of the ground clutter can be fully utilized to accurately locate the range and edge of the clutter, and provide a reference for the constant false alarm detection strategy of the clutter edge. The data analysis results prove that the range and edge of the strong sidelobe clutter can be accurately distinguished by using the technical method of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0032] Fig. 1 It is a schematic diagram of the clutter point clustering of the present application;

[0033] Fig. 2 It is a certain medium wave spectrum graph;

[0034] Fig. 3 It is an effect diagram of sidelobe clutter positioning and tracking of the present application. DETAILED DESCRIPTION

[0035] The embodiments of the present application will be described in detail below with reference to the drawings.

[0036] Following make the application's implementation through specific concrete example, the person skilled in the art can easily understand the other advantages and efficacy of the application from the disclosure of this specification. Obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments. The application can also be implemented or applied by another different specific implementation, and the details in the specification can be based on different views and applications, various modifications or changes are made without departing from the spirit of the application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor belong to the scope of protection of the present application.

[0037] It should be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms and that any particular structure and / or function described herein is merely illustrative. Based on the application, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, an apparatus and / or a method can be implemented using any number of the aspects set forth herein. In addition, this apparatus and / or method can be implemented using other structures and / or functionalities in addition to or other than one or more of the aspects set forth herein.

[0038] It should also be noted that the drawings provided in the following embodiments are only schematic and are intended to provide the basic understanding of the present application. Only the components related to the present application are shown in the drawings, not the number, shape and size of the components when actually implemented, the actual implementation of each component can be a random change, and the layout of the components can be more complex.

[0039] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0040] The embodiments of the present application provide a sidelobe clutter tracking method based on improved K-means clustering.

[0041] As Figs. 1-3 shown, the present application is directed to the performance defects of the traditional constant false alarm rate technology in the clutter edge environment of radar signal processing, and the present application provides a sidelobe clutter tracking method based on improved K-means clustering. The method uses theoretical analysis and combines a large amount of actual data to statistically analyze the range and distribution characteristics of sidelobe clutter, and provides the basis for selecting different constant false alarm detection strategies for different detection target units according to the clutter properties of different detection units.

[0042] A sidelobe clutter tracking method based on improved K-means clustering is a post-processing operation performed by a clustering method referring to data mining technology according to spatial relationship on the basis of a suspected clutter target data set formed by setting a suitable background level on a power spectrum of a traditional two-dimensional FFT spectrum, and the key lies in accurately clustering according to the distribution characteristics of the clutter patches, and the clustering result hopes to accurately condense the clutters into a class of targets, and then accurately mark which positions on the spectrum are clutter targets, thereby providing a basis for selecting different constant false alarm detection strategies, and for this purpose, a clutter clustering algorithm is designed, and the basic idea of the algorithm is: for a specific clutter target data set, the number of block clutters on the spectrum is used to initialize the classification number K in , the sample objects in the target data set and the cluster centers are calculated by iteration to obtain the minimum distance, the objects meeting the threshold are classified into a class, all classes with a final cluster center distance less than D L are merged after iteration, the clustering result is determined, and the classes with a number of targets greater than D G and a distance less than D R are reserved as clutter classes.

[0043] A sidelobe clutter tracking method based on improved K-means clustering, comprising:

[0044] Step 1: divide the RF plane into L rectangular windows of N x M, take the maximum background power position P pos (i)=[nr i ,nf i ],i=1,2,3…,L as the initialization value for subsequent clustering.

[0045] Step 2: perform sliding window processing on each detection point on the RF plane in the distance direction, and the sliding window width is W, to obtain the background P back [nr,nf] of each detection point, nr=1,2,3…,Nr,nf=1,2,3,...,Nf.

[0046] Step 3: determine whether each signal point is a suspected clutter target point, if the signal point is a suspected clutter target point, set the clutter flag of the signal point position to 1, the clutter target point Z=Z+1, otherwise set the clutter flag of the signal point position to 0, finally form a clutter binary spectrum matrix Clu[nr,nf] and store the position information S pos (i)=[nr i ,nf i ],i=1,2,3…,Z.

[0047] Step 4, count the number of clutter flag 1 in each rectangular block, if the number meets the preset threshold, it means that the rectangular block is a clutter block, and record the position of the maximum background power P in the rectangular window pos Recorded as Clu pos , record the number of rectangular windows that meet the preset threshold.

[0048] Step 5, set the input of K-means clustering algorithm, S pos As the clutter target data set, where Z is the size of the clutter data set, initialize the number of categories K in =c block And the number of iterations t of the algorithm.

[0049] Step 6, the sample object in Clu pos As the initial clustering center of K in Class.

[0050] Step 7, for each sample object S pos (i) in the clutter data set, find the clustering center Clu pos (i) with the smallest distance, if the smallest distance is less than the threshold value D L , then the sample object S pos (i) is assigned to the clustering center Clu pos (i) corresponding class C i , otherwise, the sample object is removed from the clutter data set.

[0051] Step 8, calculate the mean of each class sample object Clu pos (i) as the new clustering center point.

[0052] Step 9, calculate the clustering criterion function.

[0053] Step 10, repeat steps 7, 8, 9 until the specified iteration number t is met or the clustering criterion function converges.

[0054] The step 3 includes averaging the background power of all sampling points in each rectangular window

[0055] ,

[0056] Select the minimum background average power T back =min(P aver ) as the noise threshold for judging suspected clutter targets, initialize the number of clutter target points Z=0, multiply T back By a preset threshold coefficient K, and compare it with the background P back [nr, nf] of each detection point, if P back [nr, nf]>K×Tback If the point is considered as a suspected clutter target point.

[0057] The step 4 includes dividing into L uniform rectangular windows and initializing the number of suspected clutter blocks c block =0 on the clutter flag binary spectrum plane in the manner of step 1, summing the flag in each rectangular window and comparing with a preset threshold K flag If

[0058] ,

[0059] Record the position Clu pos of the maximum background power in the rectangular block. block (c pos (i), c block =c block +1. The preset threshold K flag is half of the number of target points in the rectangular frame.

[0060] In one embodiment, a sidelobe clutter tracking method based on improved K-means clustering further includes the following steps:

[0061] Step 11 calculates the distance between the final cluster center points of all classes C=[1,K in ] after condensation, and if the distance is less than D L , the classes are merged to form a new clutter cluster set.

[0062] Step 12 calculates the distance between the target points in each cluster in the new cluster set, and if the number of target points in each cluster is greater than G>D G and the distance between the target points in a single cluster is less than R<D R , the condensation result is taken as the clutter tracking target result data set, D G is the target number threshold value, and D R is the intra-class distance threshold value.

[0063] For a specific clutter target data set, the number of initial classification is determined according to the number of block clutter targets on the spectrum, the minimum distance between the sample objects in the target data set and the cluster center is calculated through iterative calculation, the objects meeting the threshold are classified into a class, and after the iteration is completed, all classes with a final cluster center distance less than D

[0064] In the radar air mode, the distribution characteristics of the ground clutter can be fully utilized to accurately locate the clutter range and edge, and provide a reference for the constant false alarm detection strategy of the clutter edge.

[0065] In one embodiment, a sidelobe clutter tracking method based on improved K-means clustering includes the following steps:

[0066] Step 1, after power calculation of the two-dimensional FFT spectrum, according to the distance gate number Nr and the accumulated pulse number Nf, the two-dimensional "range-frequency" (RF) plane is processed by partitioning, i.e. the RF plane is divided into L rectangular windows of N x M, and the background power of all sampling points in each rectangular window is averaged

[0067] ,

[0068] The minimum background average power T is selected from the L rectangular window average background powers back = min(P aver ) as the noise threshold for judging the suspected clutter target, and the maximum background power position P pos (i) = [nr i , nf i ], i = 1, 2, 3,..., L is taken as the subsequent clustering initialization value.

[0069] Step 2, each detection point on the RF plane is processed by sliding window in the range direction, and the sliding window width is W, to obtain the background P back [nr, nf], nr = 1, 2, 3,..., Nr, nf = 1, 2, 3,..., Nf.

[0070] Step 3, the number of clutter target points Z = 0 is initialized, T back is multiplied by a preset threshold coefficient K, and then compared with the background P back [nr, nf] of each detection point, if P back [nr, nf] > K x T back , it is considered that the point is a suspected clutter target point, the clutter flag is set to 1, the clutter target point Z = Z + 1, otherwise, the clutter flag is set to 0, and finally a clutter binary spectrum matrix Clu[nr, nf] is formed and the position information S of all suspected clutter target points is stored pos (i) = [nr i , nf i ], i = 1, 2, 3,..., Z.

[0071] Step 4, divide the clutter flag binary spectrum plane into L uniform rectangular windows and initialize the number of suspected clutter blocks c in each window as in step 1 block = 0, sum the flag in each rectangular window and compare with a pre-set threshold K flag

[0072]

[0073] record the position of the maximum background power in the rectangular block Clu pos (c block ) = P pos (i), c block = c block + 1, if c block > 0, go to step 5, otherwise output the spectrum without clutter result.

[0074] Step 5, set the input of K-means clustering algorithm, S pos as the clutter target data set, where Z is the size of the clutter data set, initialize the number of classes K in = c block and the iteration number t of the algorithm.

[0075] Step 6, take the sample objects in Clu pos as the initial clustering centers of K in classes.

[0076] Step 7, for each sample object S pos (i) in the clutter data set, find the clustering center Clu pos (i) with the minimum distance, if the minimum distance is less than the threshold value D L , then assign the sample object S pos (i) to the class C pos corresponding to the clustering center Clu i (i), otherwise, remove the sample object from the clutter data set.

[0077] Step 8, calculate the mean of the sample objects in each class as the new clustering center.

[0078] Step 9, calculate the clustering criterion function.

[0079] Step 10, repeat steps 7, 8, 9 until the specified iteration number t is met or the clustering criterion function converges.

[0080] Step 11, for the completed classes C = [1, K in ], calculate the distance between all final clustering center points, if less than D L , then merge the classes to form a new clutter clustering set. ​​

[0081] Step 12, calculate the distance between target points in a single cluster in the new cluster set, the number of targets in each cluster, and set the number of targets in each cluster G>D according to the above results G and the distance between target points in a single cluster R<D R The agglomeration results of D G are taken as the clutter tracking target result data set, D R is the target number threshold value, and D back is the intra-class distance threshold value.

[0082] In one embodiment, the radar system is X-band, Nr=366, Nf=512, the two-dimensional power spectrum is divided into 36 40x128 rectangular windows, T back =54.573db and 36 cluster center initial positions are obtained, after distance sliding window processing of each detection point of the power spectrum with a distance sliding window width W=3 2-4, K=15db, 10-20 is set, the clutter binary spectrum Clu[nr,nf] and the clutter target data set S pos are obtained, wherein the clutter data set size Z=15056, the number of suspected clutter blocks c block =12, the number of iterations t=100 80-150 of the clustering algorithm is set, D L =80, 50-100 D G =1000, 980-1020 D R =5, 4-6 The actual clutter distribution of the two-dimensional power spectrum of the embodiment is shown in Fig. 2 According to the system parameters, 36 clusters are obtained by the clutter tracking process according to the application, wherein the number of targets in a class and the distance between targets in a class of 12 classes meet D G and D R two clutter data block conditions, and the clutter tracking positioning results are shown in Fig. 3 By comparing Fig. 2 and Fig. 3 it can be concluded that the main clutter and most of the strong sidelobe clutter positions can be accurately located using the method of the application, which provides a reference basis for subsequent signal processing

[0083] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, any changes or replacements within the technical scope disclosed in the application can be easily thought of by those skilled in the art, which should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A sidelobe clutter tracking method based on improved K-means clustering, characterized in that: Comprising: Step 1, divide the RF plane into L rectangular windows Take the maximum background power position in each rectangular window as the subsequent clustering initialization value Take the maximum background power position in each rectangular window as the subsequent clustering initialization value Step 2, sliding window processing is performed on the distance direction of each detection point on the RF plane, and the sliding window width is , to obtain the background of each detection point; Step 3, judging whether each signal point is a suspected clutter target point, if the signal point is a suspected clutter target point, setting the clutter flag of the signal point position to 1, and the clutter target point , otherwise, setting the clutter flag of the signal point position to 0, and finally forming a clutter binary spectrum matrix and storing the position information of all suspected clutter target points ; Step 4, count the number of clutter flags of 1 for each rectangular block, and if the number meets a preset threshold, the rectangular block is a clutter block, and the position of the maximum background power in the rectangular window is recorded denoted as , and the number of rectangular windows meeting the preset threshold is recorded Step 5, Set the input of K-means clustering algorithm, As clutter target dataset, where For clutter dataset size, initialize the number of categories And the number of iterations of the algorithm ; Step 6, the sample objects in are taken as initial cluster centers of classes; Step 7, for each sample object in the clutter data set , find the cluster center with the minimum distance , if the minimum distance is less than a threshold , assign the sample object to the cluster center corresponding class , otherwise, remove the sample object from the clutter data set; Step 8, compute the mean of each in-class sample object as a new cluster center point; Step 9, calculating a clustering criterion function; Step 10, repeat steps 7, 8, 9 until a specified number of iterations is met or the clustering criterion function converges.

2. The improved K-means clustering based sidelobe clutter tracking method of claim 1, wherein, The step 3 includes averaging the background power of all sampling points in each rectangular window ; select the minimum background average power in L rectangular window average background power As a noise threshold to determine the suspected clutter target, initialize the number of clutter target points , multiply the preset threshold coefficient K, and compare it with the background of each detection point , if , it is considered that the point is a suspected clutter target point.

3. The improved K-means clustering based sidelobe clutter tracking method of claim 1, wherein, The step 4 includes dividing into L uniform rectangular windows and initializing the number of suspected clutter blocks on the clutter flag binary spectrum plane in the way as in step 1 , summing up the flags in each rectangular window and comparing with a preset threshold , if , the position of the maximum background power in the rectangular block is recorded .

4. The improved K-means clustering based sidelobe clutter tracking method of claim 1, wherein, Also included is step 11 to complete the clustering of the classes , calculate the distance between all final cluster centers, if less than , then merge the classes and form a new cluster set of clutter. Step 12, calculate the distance between target points in each cluster in the new cluster set, the number of targets in each cluster, and the number of targets in each cluster according to the above results and the distance between target points in each cluster The clustering result of the distance between target points in each cluster is the target number threshold value, is the intra-class distance threshold value.

5. The improved K-means clustering based sidelobe clutter tracking method of claim 3, wherein, The preset threshold is half of the number of target points in the rectangular frame.

Images