A Constant False Alarm Rate Detection Method Based on Background Discrimination

Abstract

The present invention belongs to the field of radar constant false alarm detection and relates to a constant false alarm detection method based on background discrimination. The method of the present invention adopts the idea of deleting constant false alarm detection, ensuring the detection probability in the case of target occlusion, so that the detection probability of the present invention is higher in the case of target occlusion. The present invention calculates the clutter separation point within the full detection range, further improving the stability of the false alarm probability near the clutter demarcation point and enhancing the detection performance. By adopting different detection strategies in different situations, the method of the present invention can achieve target detection under multi-target interference and improve the stability of the false alarm probability at the clutter edge.

Description

Technical Field

[0001] The present invention belongs to the field of radar constant false alarm detection, and relates to a constant false alarm detection method based on background discrimination. Background Art

[0002] Constant False Alarm Rate (CFAR) detection is an important part of radar signal processing. It can adjust the detection threshold according to the change of the detection background to ensure a certain false alarm probability. Therefore, the research on CFAR detection algorithms has received widespread attention. The five commonly used CFAR detection methods are: Cell-Averaging CFAR (CA-CFAR), Greatest-of CFAR (GO-CFAR), Smallest-of CFAR (SO-CFAR), Ordered Statistics CFAR (OS-CFAR), and Adaptive CFAR. Among them, the cell mean type CFAR is only applicable to a single situation and cannot meet the requirements of multi-target and clutter edge situations at the same time. VI-CFAR can adaptively select the cell mean type CFAR according to different backgrounds, but its detection performance will decline in a multi-target environment.

[0003] Smith et al. introduced a CFAR detector based on variability index (VI-CFAR) in the literature Intelligent CFAR processor based on data variability. The strategy of this detector is to select different mean type CFARs for target detection. However, its detection performance will decline in an environment where interference targets exist simultaneously in the reference windows on both sides of the cell to be detected.

[0004] Finn proposed a non-uniform clutter estimation CFAR processing method (HCE-CFAR) in the literature A CFAR Design for a Window Spanning Two Clutter Fields. It is an adaptive CFAR method applicable to non-uniform backgrounds. This method first finds the critical cells in different backgrounds, determines which background area the detection cell is located in, then evaluates the background power of this area, and finally performs CFAR detection. The stability of the false alarm rate near the clutter boundary point of this method has been improved, but its detection performance will decline in the case of multi-target interference. Summary of the Invention

[0005] In order to solve the problem of missed detection in the case of multi-targets existing in constant false alarm detection and improve the stability of the false alarm probability at the clutter edge, the present invention proposes a constant false alarm detection method based on background discrimination, which improves both the detection probability in a multi-target environment and the stability of the false alarm probability at the clutter edge.

[0006] The present invention utilizes maximum likelihood estimation. Maximum likelihood estimation is an important and common method for obtaining estimators. The maximum likelihood method explicitly uses a probability model, and its goal is to find a phylogenetic tree that can generate the observed data with a relatively high probability. Compared with other estimation methods, the performance of maximum likelihood estimation is more stable. It is used to calculate the position of the clutter separation point in the present invention.

[0007] The mathematical formula derivation of maximum likelihood estimation is as follows:

[0008] Given a probability distribution D, assuming its probability density function is f D , and a distribution parameter θ, we can draw a sample x1,..., x n with n values from this distribution. By using f D , we can calculate its probability and then write out its likelihood function

[0009]

[0010] where, ∏ represents the product operation, p(x i ; θ) represents the probability that X = x, represents the event

[0011] {X1 = x1,..., Xn = xn} occurs.

[0012] Take the logarithm of both sides to get

[0013]

[0014] where, ∑ represents the summation operation, and ln represents the logarithm with the constant e as the base

[0015] Take the derivative of lnL(θ) and set it to 0

[0016]

[0017] where represents the derivative of the function lnL(θ) with respect to θ

[0018] This equation is the log-likelihood equation. The solution of the log-likelihood equation is the maximum likelihood estimate value of the unknown parameter.

[0019] The technical solution of the present invention is as follows:

[0020] S1. For the detection sequence (x1,..., x N ) (the sampling values after the echo signal passes through square-law detection), determine whether the detection background (the reference cells included in the reference window, see Figure 2 ) is uniform. If so, go to step S2; otherwise, go to S4;

[0021] S2. Calculate the detection threshold, specifically as follows:

[0022] Define the reference cell sequence as (x1,..., x M ), and the reference cells are the detection sequences on both sides of the cell to be detected (i.e., an element in the detection sequence (x1,..., x N ) used to calculate the threshold. Define that the number of reference cells in both the leading reference window and the trailing reference window is M / 2, and if the probability density function distribution of the reference cell sequence (x1,..., x M ) follows an exponential distribution, then for k = 1,..., M - 1, the likelihood function expressions are obtained respectively:

[0023]

[0024] where represents the background power on both sides of K. Take the logarithm of both sides of the above formula, and respectively let the result of taking the derivative of ln(L(K)) with respect to be 0. The estimation results for k = 1,..., M - 1 are equivalent to;

[0025]

[0026] where represents the value of K when takes the minimum value. Then the detection threshold T is obtained as:

[0027]

[0028] S3. Determine whether there is a cell x m in the detection sequence that exceeds the detection threshold T. If so, let:

[0029]

[0030] And recalculate the detection threshold T according to the method in S2 within the range (m - M / 2,..., m + M / 2) of the detection sequence and repeat S3; otherwise, go to S6;

[0031] S4. Determine the position of the clutter edge throughout the detection range through the following formula:

[0032]

[0033] where,

[0034] According to the positions of the clutter separation points with different background powers obtained Set M protection units on both the left and right sides (since the power of the target may leak into adjacent units, several units adjacent to the target are not used for estimating the background clutter and are used as protection units). p Each has M. The protection units are detection sequences adjacent to the unit to be detected and are located between the reference unit and the unit to be detected; determine whether the reference unit and the protection units contain clutter separation points. If so, enter S5; otherwise, enter S2;

[0035] S5. Determine whether the unit to be detected is in a clutter area with a high power background. If so, calculate the detection threshold:

[0036]

[0037] After obtaining the detection threshold, enter S6; otherwise, enter S2;

[0038] S6. Determine whether the detection sequence is greater than the detection threshold. If so, determine that there is a target; otherwise, determine that there is no target.

[0039] Furthermore, the specific method for determining whether the detection background is uniform in S1 is as follows:

[0040] Define the position of the clutter separation point as k. When k = 1,..., N - 1, calculate the background means before and after the separation point k respectively:

[0041]

[0042] Calculate the average power of the detection sequence:

[0043]

[0044] Perform background discrimination:

[0045]

[0046] Where ΔW represents the division interval of the background power. If the background powers differ by more than ΔW, it indicates that the detection background is divided into clutter backgrounds with different powers.

[0047] The beneficial effects of the present invention are as follows:

[0048] 1) The method of the present invention adopts the idea of deleted constant false alarm detection, ensuring the detection probability in the case of target occlusion, making the detection probability of the present invention higher in the case of target occlusion;

[0049] 2) The present invention calculates the clutter separation point within the entire detection range, further improving the stability of the false alarm probability near the clutter demarcation point and improving the detection performance.

[0050] 3) Different detection strategies are adopted in different situations. The method of the present invention can achieve target detection under multi-target interference and improve the stability of the false alarm probability at the clutter edge. Description of the Drawings

[0051] Figure 1 It is a flowchart of the present invention.

[0052] Figure 2 It is a schematic diagram of the detection reference window.

[0053] Figure 3 They are the detection probabilities of each method in a multi-target environment.

[0054] Figure 4 They are the false alarm performances of each method in a clutter edge environment. Detailed Embodiment

[0055] The technical solution of the present invention will be described in detail below in conjunction with the drawings and embodiments:

[0056] Since the method of the present invention is applied to target detection, an application example of the method of the present invention in target detection will be described below.

[0057] Embodiment

[0058] In this example, the false alarm probability P fa = 10 -4 , the length of the reference window is 20, the number of left and right protection units Mp = 1, and it is set that the background is non-uniform when the difference in background power is greater than 5 dB, that is, ΔW = 5.

[0059] 1) Randomly generate 400 detection sequences subject to Gaussian distribution. The first 100 detection sequences are subject to Gaussian distribution with a power of 20 dB, and the last 100 detection sequences are subject to Gaussian distribution with a power of 30 dB. A target with a signal-to-noise ratio of 15 dB is added to the 57th, 61st, and 65th units respectively.

[0060] 2) According to formula (4), when k = 1,..., N - 1, calculate the background means before and after the kth point respectively, and the minimum and maximum values of M(k) are 0.6973 and 0.9110 respectively;

[0061] 3) According to formula (6), W1 = 0.8189 and W2 = 1.1811 are obtained. There are values of M(k) that do not conform to formula (6), and it is judged that the background is non-uniform;

[0062] 4) Execute step d. Substitute the entire detection sequence into formula (10) to calculate and obtain

[0063] 5) According to the formula (11) and the result of 4), when the detection unit sequence is less than or equal to 99, calculate the detection threshold according to the formulas (7) and (8). The threshold results of the 57th, 61st, and 65th units are 37.9026, 35.941, and 29.8892 respectively. Among them, the 57th and 65th units exceed the threshold. According to the formula (9), replace the values of the 57th and 65th units, and then repeat 5) until no new unit to be detected exceeds the threshold.

[0064] 6) When the detection unit sequence is greater than 99 and less than 121, calculate the threshold according to the formula (12).

[0065] 7) Compare the detection threshold with the sequence to be detected to determine whether there is a target.

[0066] Contents of Monte Carlo experiment:

[0067] Target detection is carried out in a multi-target environment and a clutter edge environment. The comparison methods are CA-CFAR, SO-CFAR, GO-CFAR, and HCE-CFAR. The parameter settings of the comparison methods are all: there is 1 guard cell and 20 reference cells at the front and rear edges of the detection unit. This method sets 1 guard cell and 20 reference cells for each of the front and rear windows, does not set a guard cell, and ΔW = 5.

[0068] In the comparison experiment in a multi-target environment, SNR = 20dB, and the targets are at range cell 101, range cell 97, and range cell 105, and the signal-to-noise ratios are all from 0dB to 30dB. From the simulation comparison Figure 2 it can be seen that at the same signal-to-noise ratio, the detection probability of this method is higher than that of the other four methods.

[0069] In the comparison experiment in a clutter edge environment, SNR = 20dB for the first 100 range cells and SNR = 30dB for the last 100 range cells. From the simulation comparison Figure 3 it can be seen that the stability of the false alarm probability of this method in a clutter edge environment is better than that of the other four CFARs.

[0070] The present invention mainly combines the idea of the change index and non-uniform clutter CFAR, and uses maximum likelihood estimation in the estimation process to achieve more stable detection performance at the multi-target interference and clutter boundary points.

Claims

1. A constant false alarm detection method based on background discrimination, characterized in that, It includes the following steps: S1. For a given detection sequence (x1,...,x N ), determine whether the detection background is uniform. If so, proceed to step S2; otherwise, proceed to S4. S2. Calculate the detection threshold, specifically: Define the reference unit sequence as (x1,..., x M ), where the reference units are the detection sequences used to calculate the threshold on both sides of the unit to be detected. The unit to be detected refers to an element in the detection sequence. Define that the number of reference units in both the leading reference window and the trailing reference window is M / 2, and if the probability density function distribution of the reference unit sequence (x1,..., x M ) follows an exponential distribution, then for k = 1,..., M - 1, the likelihood function expressions are obtained respectively: Among them represents the background power on both sides of K. Take the logarithm of both sides of the above formula, and let the result of taking the derivative of ln(L(K)) with respect to be 0. The estimation results for k = 1,..., M - 1 are equivalent to; Among them indicates that when making take the minimum value of K, the obtained detection threshold T is as follows: where P fa is the false alarm probability; S3. Determine whether there is a cell x in the detection sequence that exceeds the detection threshold T m , if so, then let: And within the range of (m - M / 2,..., m + M / 2) of the detection sequence, recalculate the detection threshold T according to the method of S2 and repeat the execution of S3; otherwise, enter S6; S4. Determine the position of the clutter edge within the entire detection range through the following formula: Among them, According to the positions of the obtained different background power clutter separation points Set M for each of the left and right protection units p pieces. The protection unit is a detection sequence adjacent to the unit to be detected and is located between the reference unit and the unit to be detected; determine whether the reference unit and the protection unit contain clutter separation points. If so, enter S5; otherwise, enter S2; S5. Determine whether the unit to be detected is in a clutter area with a relatively high power background. If so, calculate the detection threshold: After obtaining the detection threshold, enter S6; otherwise, enter S2; S6. Determine whether the detection sequence is greater than the detection threshold. If so, it is determined that there is a target; otherwise, it is determined that there is no target.

2. The constant false alarm detection method based on background discrimination according to claim 1, characterized in that, The specific method for determining whether the detection background is uniform in S1 is: Define the position of the clutter separation point as k. When k = 1,..., N - 1, calculate the background means before and after the separation point k respectively: Calculate the average power of the detection sequence: Perform discrimination through the background discrimination formula: Where ΔW represents the division interval of the background power. If the difference in background power is greater than ΔW, it indicates that the detection background is divided into clutter backgrounds with different powers. If the background discrimination formula holds, the detection background is a uniform environment; Otherwise, the detection background is a non-uniform environment.

Images