CN117420521A - Self-adaptive multi-target CFAR detection method based on local outlier factors - Google Patents

Abstract

The invention relates to a self-adaptive multi-target CFAR detection method based on local outlier factors, and belongs to the technical field of radar signal processing. Aiming at CFAR detection of an actual radar system in a complex multi-target environment, the method firstly utilizes a local outlier factor algorithm to perform clustering processing on echo signals in a front sliding window and a rear sliding window, and eliminates potential abnormal points such as interference targets; then, estimating the background power value of the unit to be detected according to the reference unit signal after the abnormal points are removed; and finally, judging the real power value of the unit to be detected and the detection threshold value by using a comparator to obtain a final CFAR detection result. The invention can cluster the data points in the reference sliding window into one or more clusters with arbitrary shapes, and the prior information of the distribution condition of the targets and the threshold parameters of the number of the targets are not needed, and the number of the clusters is not needed to be declared in advance; the method can effectively inhibit the 'target shielding' effect and realize the self-adaptive CFAR detection of multiple targets.

Description

Self-adaptive multi-target CFAR detection method based on local outlier factors

Technical Field

The invention belongs to the technical field of radar signal processing, and relates to a self-adaptive multi-target constant false alarm rate (Constant False Alarm Rate, CFAR) detection method based on local outlier factors (Local Outlier Factor, LOF), which is suitable for adaptively detecting targets in a multi-target environment.

Background

CFAR detection is an important technology for radar to automatically detect targets, and can maximize detection probability while maintaining constant false alarm probability. CFAR detection is typically performed in noisy and noisy environments, particularly in multi-target environments, when one or more interfering targets are simultaneously located in either the leading or trailing edge sliding windows, a severe "target masking" effect is created, causing target omission, thus degrading detection performance.

Currently, commonly used CFAR detectors include a cell average CFAR detector (CA-CFAR), a minimum selected CFAR detector (SO-CFAR), a maximum selected CFAR detector (GO-CFAR), an ordered statistics CFAR detector (OS-CFAR), and a deleted mean CFAR detector (CMLD-CFAR). Their drawbacks are represented by: (1) CA-CFAR and GO-CFAR detectors are applicable to simple environments such as uniform background and clutter edges, respectively, but are not applicable to complex multi-target environments; (2) SO-CFAR, OS-CFAR and CMLD-CFAR detectors can be used in a multi-target environment, but they need to know the distribution of different targets in advance and set threshold parameters for the number of targets artificially, otherwise the detection performance will be significantly deteriorated. However, most practical application scenarios are multi-target environments, and prior information such as distribution conditions of targets and the number of targets is often difficult to accurately acquire in advance.

Therefore, the invention designs the CFAR detection method which is suitable for the multi-target environment, does not need prior information of the target distribution condition and does not need preset target number threshold parameters.

Disclosure of Invention

The invention aims at solving the defects that the prior information of the target distribution condition is difficult to obtain under the complex multi-target clutter background environment of the existing CFAR detector and the target number threshold parameters are manually set, and provides a local outlier factor-based self-adaptive multi-target CFAR detection method, namely LOF-CFAR.

The invention adopts the solution scheme for solving the technical problems that: firstly, clustering echo signals in a front sliding window and a back sliding window by utilizing a local outlier factor algorithm, and removing abnormal points such as potential interference targets positioned in the front sliding window and the back sliding window; then, estimating the background power value of the unit to be detected according to the reference unit signal after the abnormal points are removed; and finally, judging the real power value of the unit to be detected and the detection threshold value by using a comparator to obtain a final CFAR detection result.

The method comprises the following specific steps:

s1: receiving radar echo signals and carrying out matched filtering treatment;

s2: setting the reference sliding window of the unit to be tested as N=2n (namely the front sliding window and the rear sliding window are both N), and setting the number of the protection units as M=2m (namely the number of the protection units positioned on the left side and the right side of the unit to be tested is M);

s3: echo signal x= [ x ] after matched filtering _1I +jx _1Q ,x _2I +jx _2Q ,…,x _NI +jx _NQ ]Loading a reference sliding window, extracting in-phase components and quadrature components of signals corresponding to each reference unit to form a two-dimensional data point set D= { (x) _1I ,x _1Q ),(x _2I ,x _2Q ),…,(x _NI ,x _NQ )}；

S4: clustering two-dimensional data points in the set D by utilizing a local outlier factor algorithm to form one or more clusters with arbitrary shapes, and separating out normal points and abnormal points according to a clustering result, wherein a reference unit corresponding to the abnormal points is s, and s is E [1, N ];

s5: processing the echo signals after matched filtering in a square rate detection mode to obtain corresponding signals of each reference unit asWhere i=1, 2, …, N;

s6: removing the corresponding abnormal signal x obtained in the step S5 according to the reference unit S corresponding to the abnormal point obtained in the step S4 _s ；

S7: for the corresponding signal { x } of the remaining reference cells in the reference sliding window after the outlier is removed _i \x _s Performing average treatment, and estimating to obtain a background power value Z of the unit to be detected;

s8: calculating a detection threshold value T according to the estimated background power value Z and a threshold factor;

s9: real power value x of unit to be tested by using comparator ₀ With detection thresholdJudging the threshold T to obtain a final CFAR detection result H ₁ Or H ₀ 。

Further, the specific process of clustering the two-dimensional data points in the set D in the step S4 is as follows:

let p and q be any two points in the set D, the euclidean distance between them being denoted D (p, q). The distance from the kth nearest neighbor node defining point p to point p is referred to as the k-distance of point p, denoted d _k (p), wherein k is a natural number. The set of the rest of points except the point p in the set D is defined as d\ { p }. If at least k different points q E D\ { p } satisfy D (p, q) < D _k (p) and at most k-1 points q εD\ { p } satisfy D (p, q)<d _k (p), then d (p, q) is equivalent to d _k (p)。

For a given d _k (p) defining that all distances to point p in set D are less than D _k The neighborhood formed by the points of (p) is referred to as the k-distance neighborhood of point p, namely:

N _k (p)＝{q∈D\{p}∣d(p,q)≤d _k (p)} (1)

and |N _k (p) | represents the number of all points contained within the k-distance neighborhood of point p.

Furthermore, the maximum value of the k-distance of the point p and the euclidean distance between the points p and q is defined as the reachable distance from the point p to the point q, namely:

d _rk (p,q)＝max{d _k (p),d(p,q)} (2)

meanwhile, taking the reciprocal of the average value of all the reachable distances from the point p to the point q in the k-distance neighborhood of the point p, namely:

wherein d _lrk (p) represents the locally reachable density of points p.

Definition of local outlier factor LOF _k (p) is the density difference of point p from the overall data point, namely:

thus, if the local outlier factor LOF _k The value of (p) is far greater than 1, which indicates that the local reachable density around the point p has large density difference from the whole, and the point p is considered to be an abnormal point; if local outlier factor LOF _k The value of (p) is close to 1, indicating that the local reachable density around point p differs less from the overall density, point p is considered to be a normal point.

Based on the method, the cyclic process of eliminating the abnormal points such as the interference targets in the reference sliding window by utilizing the local outlier factor algorithm comprises the following steps:

s4-1: for point p ε D, calculate Euclidean distance D (p, q) between point p and the rest of points q in set D;

s4-2: determining the k-distance d of point p _k (p) and the number of all points contained in the k-distance neighborhood of point p |N _k (p)|；

S4-3: calculating the reachable distance d from the point p to the point q by the formula (2) _rk (p,q)；

S4-4: calculating the local reachable density d of the point p by the formula (3) _lrk (p)；

S4-5: calculating local outlier factor LOF at point p by equation (4) _k (p)；

S4-6: judging whether the point p is an abnormal point or not;

s4-7: go to step S4-1 until all points in set D are traversed.

Further, the specific process of estimating the background power value Z of the unit under test in step S7 is as follows:

wherein i=1, 2, …, N, S e [1, N ], S represents the number of outliers in the current reference sliding window.

Further, the specific process of calculating the detection threshold T in step S8 is as follows:

T＝αZ (6)

wherein alpha represents a threshold factor, Z represents an estimated background power value of the unit to be detected, and the acquisition of Z is described in formula (5).

Further, the specific process of determining the final CFAR detection result in step S9 is:

wherein H is ₁ Indicating the presence of a target in the unit under test, H ₀ Indicating that no target is present in the unit under test.

Compared with the prior art, the invention has the remarkable advantages that: aiming at the requirement of multi-target detection of an actual radar system in a complex environment, the invention designs a new LOF-CFAR detector by combining a local outlier factor algorithm, and can cluster data points in a reference sliding window into one or more clusters with arbitrary shapes, without prior information of target distribution conditions and threshold parameters of target number and without prior statement of the number of clusters; by separating the abnormal point from the normal point and eliminating potential interference targets located on the front sliding window and the rear sliding window, adverse effects of the target shielding effect on detection results can be effectively restrained, and multi-target self-adaptive CFAR detection is realized.

Drawings

FIG. 1 is a basic block diagram of a LOF-CFAR detector;

FIG. 2 is a graph showing comparison of detection performance of different CFAR detectors with a random interference target number of 1;

FIG. 3 is a graph showing comparison of detection performance of different CFAR detectors with a random interference target number of 2;

FIG. 4 is a graph showing comparison of detection performance of different CFAR detectors with a random interference target number of 4;

fig. 5 is a graph comparing detection performance of different CFAR detectors with a random interference target number of 5.

Detailed Description

A specific embodiment of the present invention will be described in detail with reference to fig. 1. An adaptive multi-target CFAR detector based on local outliers according to the present embodiment includes the steps of:

s1: receiving radar echo signals and carrying out matched filtering treatment;

s2: setting the reference sliding window of the unit to be tested as N=2n (namely the front sliding window and the rear sliding window are both N), and setting the number of the protection units as M=2m (namely the number of the protection units positioned on the left side and the right side of the unit to be tested is M);

s3: echo signal x= [ x ] after matched filtering _1I +jx _1Q ,x _2I +jx _2Q ,…,x _NI +jx _NQ ]Loading a reference sliding window, extracting in-phase components and quadrature components of signals corresponding to each reference unit to form a two-dimensional data point set D= { (x) _1I ,x _1Q ),(x _2I ,x _2Q ),…,(x _NI ,x _NQ )}；

S4: clustering two-dimensional data points in the set D by utilizing a local outlier factor algorithm to form one or more clusters with arbitrary shapes, and separating out normal points and abnormal points according to a clustering result, wherein a reference unit corresponding to the abnormal points is s, and s is E [1, N ];

s5: processing the echo signals after matched filtering in a square rate detection mode to obtain corresponding signals of each reference unit asWhere i=1, 2, …, N;

s6: removing the corresponding abnormal signal x obtained in the step S5 according to the reference unit S corresponding to the abnormal point obtained in the step S4 _s ；

S7: for the corresponding signal { x } of the remaining reference cells in the reference sliding window after the outlier is removed _i \x _s Performing average treatment, and estimating to obtain a background power value Z of the unit to be detected;

s8: calculating a detection threshold value T according to the estimated background power value Z and a threshold factor;

s9: real power value x of unit to be tested by using comparator ₀ And a detection threshold valueT makes a decision to obtain a final CFAR detection result H ₁ Or H ₀ 。

Further, the specific process of clustering the two-dimensional data points in the set D in the step S4 is as follows:

let p and q be any two points in the set D, the euclidean distance between them being denoted D (p, q). The distance from the kth nearest neighbor node defining point p to point p is referred to as the k-distance of point p, denoted d _k (p), wherein k is a natural number. The set of the rest of points except the point p in the set D is defined as d\ { p }. If at least k different points q E D\ { p } satisfy D (p, q) < D _k (p) and at most k-1 points q εD\ { p } satisfy D (p, q)<d _k (p), then d (p, q) is equivalent to d _k (p)。

For a given d _k (p) defining that all distances to point p in set D are less than D _k The neighborhood formed by the points of (p) is referred to as the k-distance neighborhood of point p, namely:

N _k (p)＝{q∈D\{p}∣d(p,q)≤d _k (p)} (1)

and |N _k (p) | represents the number of all points contained within the k-distance neighborhood of point p.

Furthermore, the maximum value of the k-distance of the point p and the euclidean distance between the points p and q is defined as the reachable distance from the point p to the point q, namely:

d _rk (p,q)＝max{d _k (p),d(p,q)} (2)

meanwhile, taking the reciprocal of the average value of all the reachable distances from the point p to the point q in the k-distance neighborhood of the point p, namely:

wherein d _lrk (p) represents the locally reachable density of points p.

Definition of local outlier factor LOF _k (p) is the density difference of point p from the overall data point, namely:

thus, if the local outlier factor LOF _k The value of (p) is far greater than 1, which indicates that the local reachable density around the point p has large density difference from the whole, and the point p is considered to be an abnormal point; if local outlier factor LOF _k The value of (p) is close to 1, indicating that the local reachable density around point p differs less from the overall density, point p is considered to be a normal point.

Based on the method, the cyclic process of eliminating the abnormal points such as the interference targets in the reference sliding window by utilizing the local outlier factor algorithm comprises the following steps:

s4-1: for point p ε D, calculate Euclidean distance D (p, q) between point p and the rest of points q in set D;

s4-2: determining the k-distance d of point p _k (p) and the number of all points contained in the k-distance neighborhood of point p |N _k (p)|；

S4-3: calculating the reachable distance d from the point p to the point q by the formula (2) _rk (p,q)；

S4-4: calculating the local reachable density d of the point p by the formula (3) _lrk (p)；

S4-5: calculating local outlier factor LOF at point p by equation (4) _k (p)；

S4-6: judging whether the point p is an abnormal point or not;

s4-7: go to step S4-1 until all points in set D are traversed.

Further, the specific process of estimating the background power value Z of the unit under test in step S7 is as follows:

wherein i=1, 2, …, N, S e [1, N ], S represents the number of outliers in the current reference sliding window.

Further, the specific process of calculating the detection threshold T in step S8 is as follows:

T＝αZ (6)

wherein alpha represents a threshold factor, Z represents an estimated background power value of the unit to be detected, and the acquisition of Z is described in formula (5).

Further, the specific process of determining the final CFAR detection result in step S9 is:

wherein H is ₁ Indicating the presence of a target in the unit under test, H ₀ Indicating that no target is present in the unit under test.

The detection performance of the invention is further verified and illustrated by the following simulation experiment.

Simulation parameter setting: assuming that the number of distance units of the radar echo signal is 500, the numbers of reference units and protection units are n=64 and m=4, respectively. In contrast to the present invention, the method comprises CA-CFAR, SO-CFAR, GO-CFAR, OS-CFAR and CMLD-CFAR detectors, wherein the CA-CFAR, SO-CFAR and GO-CFAR detectors have no additional configuration parameters, and the configuration parameters of the OS-CFAR, CMLD-CFAR and LOF-CFAR detectors are t=60, r=1 and k=10, respectively. The false alarm rate is set to P _fa ＝10 ^-4 The Monte Carlo simulation times are 10000 times. Four typical multi-target scenes are sequentially arranged, and each scene comprises 1,2, 4 and 5 random interference targets, so that the detection performance of the designed method and the detection performance of the compared method are verified; the detection probability is used as a performance index, and the experimental results are shown in fig. 2 to 5.

FIG. 2 is a comparison of detection performance of different CFAR detectors with a random interference target number of 1; FIG. 3 is a comparison of detection performance of different CFAR detectors with a random interference target number of 2; FIG. 4 is a comparison of detection performance of different CFAR detectors with a random interference target number of 4; fig. 5 is a comparison result of detection performance of different CFAR detectors in the case that the number of random interference targets is 5.

When there is only one random interference target in the reference sliding window, the LOF-CFAR, SO-CFAR, OS-CFAR and CMLD-CFAR detectors can all conveniently detect a single interference target. The detection probability of these detectors is greater than 62.5% when the SCR value is 18 dB. However, the CA-CFAR detector ignores the influence of the interference target, so that the detection threshold is too high, and the detection performance of the CA-CFAR detector is poor under the condition of low signal-to-noise ratio. Furthermore, the GO-CFAR detector has the worst detection performance due to the severe "target shielding" effect.

The detection performance of SO-CFAR and CMLD-CFAR is significantly degraded when the number of random interference targets in the reference sliding window is increased to 2 and 4. At this point, the number of interference targets has exceeded the tolerance of the CMLD-CFAR detector, which can only tolerate one interference target at most, since the parameter is configured to r=1.

When the random interference targets in the reference sliding window are 5, the performance advantage of the OS-CFAR detector is no longer present, since the number of interference targets at this time has exceeded the preconfigured parameter N-t=4. However, even in the case where the number of interference targets increases, the inventive LOF-CFAR detector can always maintain good detection performance. Furthermore, the LOF-CFAR detector has no special requirement on the tolerance of the number of interference targets, and no prior information is required on the distribution of multiple interference targets in the front-edge sliding window or the back-edge sliding window.

Claims (5)

1. An adaptive multi-target CFAR detection method (namely LOF-CFAR) based on local outlier factors is characterized by comprising the following steps:

s1: receiving radar echo signals and carrying out matched filtering treatment;

s2: setting the reference sliding window of the unit to be tested as N=2n (namely the front sliding window and the rear sliding window are both N), and setting the number of the protection units as M=2m (namely the number of the protection units positioned on the left side and the right side of the unit to be tested is M);

s3: echo signal x= [ x ] after matched filtering _1I +jx _1Q ,x _2I +jx _2Q ,…,x _NI +jx _NQ ]Loading a reference sliding window, extracting in-phase components and quadrature components of signals corresponding to each reference unit to form a two-dimensional data point set D= { (x) _1I ,x _1Q ),(x _2I ,x _2Q ),…,(x _NI ,x _NQ )}；

S4: clustering two-dimensional data points in the set D by utilizing a local outlier factor algorithm to form one or more clusters with arbitrary shapes, and separating out normal points and abnormal points according to a clustering result, wherein a reference unit corresponding to the abnormal points is s, and s is E [1, N ];

s5: processing the echo signals after matched filtering in a square rate detection mode to obtain corresponding signals of each reference unit asWhere i=1, 2, …, N;

s6: removing the corresponding abnormal signal x obtained in the step S5 according to the reference unit S corresponding to the abnormal point obtained in the step S4 _s ；

S7: for the corresponding signal { x } of the remaining reference cells in the reference sliding window after the outlier is removed _i \x _s Performing average treatment, and estimating to obtain a background power value Z of the unit to be detected;

s8: calculating a detection threshold value T according to the estimated background power value Z and a threshold factor;

s9: real power value x of unit to be tested by using comparator ₀ Judging with a detection threshold value T to obtain a final CFAR detection result H ₁ Or H ₀ 。

2. The method for adaptive multi-objective CFAR detection based on local outliers according to claim 1, wherein the specific process of clustering the two-dimensional data points in the set D in step S4 is as follows:

assuming that p and q are any two points in set D, the euclidean distance between them is denoted as D (p, q); the distance from the kth nearest neighbor node defining point p to point p is referred to as the k-distance of point p, denoted d _k (p), wherein k is a natural number; defining the rest points except the point p in the set D as D\ { p }; if at least k different points q E D\ { p } satisfy D (p, q) < D _k (p), and toThere are k-1 more points q εD\ { p } satisfying D (p, q)<d _k (p), then d (p, q) is equivalent to d _k (p); for a given d _k (p) defining that all distances to point p in set D are less than D _k The neighborhood formed by the points of (p) is referred to as the k-distance neighborhood of point p, namely:

N _k (p)＝{q∈D\{p}∣d(p,q)≤d _k (p)} (1)

and |N _k (p) | represents the number of all points contained within the k-distance neighborhood of point p; furthermore, the maximum value of the k-distance of the point p and the euclidean distance between the points p and q is defined as the reachable distance from the point p to the point q, namely:

d _rk (p,q)＝max{d _k (p),d(p,q)} (2)

meanwhile, taking the reciprocal of the average value of all the reachable distances from the point p to the point q in the k-distance neighborhood of the point p, namely:

wherein d _lrk (p) represents the locally reachable density of points p; definition of local outlier factor LOF _k (p) is the density difference of point p from the overall data point, namely:

thus, if the local outlier factor LOF _k The value of (p) is far greater than 1, which indicates that the local reachable density around the point p has large density difference from the whole, and the point p is considered to be an abnormal point; if local outlier factor LOF _k A value of (p) close to 1 indicates that the local reachable density around point p differs less from the overall density, point p being considered a normal point; based on the method, the cyclic process of eliminating the abnormal points such as the interference targets in the reference sliding window by utilizing the local outlier factor algorithm comprises the following steps:

s4-1: for point p ε D, calculate Euclidean distance D (p, q) between point p and the rest of points q in set D;

s4-2: determining the k-distance d of point p _k (p) and the number of all points contained in the k-distance neighborhood of point p |N _k (p)|；

S4-3: calculating the reachable distance d from the point p to the point q by the formula (2) _rk (p,q)；

S4-4: calculating the local reachable density d of the point p by the formula (3) _lrk (p)；

S4-5: calculating local outlier factor LOF at point p by equation (4) _k (p)；

S4-6: judging whether the point p is an abnormal point or not;

s4-7: go to step S4-1 until all points in set D are traversed.

3. The method for adaptive multi-objective CFAR detection based on local outliers according to claim 1, wherein the specific process of estimating the background power value Z of the unit under test in step S7 is as follows:

wherein i=1, 2, …, N, S e [1, N ], S represents the number of outliers in the current reference sliding window.

4. The method for adaptive multi-objective CFAR detection based on local outliers according to claim 1, wherein the specific process of calculating the detection threshold T in step S8 is:

T＝αZ (6)

wherein alpha represents a threshold factor, Z represents an estimated background power value of the unit to be detected, and the acquisition of Z is described in formula (5).

5. The adaptive multi-objective CFAR detection method according to claim 1, wherein the specific process of determining the final CFAR detection result in step S9 is as follows:

wherein H is ₁ Indicating the presence of a target in the unit under test, H ₀ Indicating that no target is present in the unit under test.