CN105572651A - CFAR detection method based on clutter background statistical recognition - Google Patents

Abstract

A CFAR detection method based on clutter background statistical recognition, the invention relates to a CFAR detection method based on clutter background statistical recognition. The purpose of the present invention is to solve the problem that when the existing actual detection background statistics no longer satisfy the Gaussian background, the performance of the detector will decrease and the actual detection background is no longer uniformly distributed, causing the false alarm rate to increase or the detection probability to decrease. The specific process is: 1. Start; 2. Input the RD spectrum of the data; 3. Perform KL divergence partition on the RD spectrum in 2 to obtain the partitioned data; 4. Perform parameter estimation on the partitioned data to obtain the estimated parameters 5. Using the estimated parameters to normalize the background into an exponential distribution to obtain a normalized detection background; 6. Perform CFAR detection on the normalized detection background to obtain a CFAR detection result. The invention is applied in the field of target detection and processing under complex clutter background.

Description

A CFAR Detection Method Based on Statistical Recognition of Clutter Background

technical field

The invention relates to a CFAR detection method based on clutter background statistical recognition.

Background technique

Traditional CFAR detection is based on the assumption of Gaussian background and clutter statistical independence. The problem is that when the actual detection background statistics no longer satisfy the Gaussian background, the performance of the detector will drop sharply.

CFAR algorithms based on Gaussian background are divided into two categories: mean level (ML) CFAR detectors and ordered statistics (OS) CFAR detectors. Among them, the average constant false alarm detector includes CA, GO, and OS-CFAR detection algorithms. In uniform clutter backgrounds, CA-CFAR achieves optimal detection, but its performance degrades in non-uniform environments. GO-CFAR can improve the detection performance in the edge case of clutter, and SO-CFAR has better anti-interference ability. In the multi-target environment, the detection performance of the OS class detection algorithm is better than that of the mean class CFAR. The traditional CFAR detection background is based on the assumption that the Gaussian background and clutter statistics are independent. There are two problems in the traditional CFAR algorithm: one is that the actual detection background distribution no longer satisfies the exponential distribution (the echo obeys Gaussian distribution, the output of the envelope detector is Rayleigh distribution, and the output of square law detection is exponential distribution), and the clutter The mismatch between the actual distribution and the design model of the detector, if the traditional CFAR detector is used, the detection performance of the detector will decrease; the second is that the actual detection background is no longer a uniform distribution, and the background distribution is complex. When detecting targets in this background, it is inevitable It will bring performance loss, cause the false alarm rate to increase or the detection probability to decrease. From the perspective of the radar system, it is often hoped that the actual false alarm probability is near the set value, and the false alarm probability does not want to change drastically.

Contents of the invention

The purpose of the present invention is to solve the problem that when the existing actual detection background statistics no longer satisfy the Gaussian background, the performance of the detector will decrease and the actual detection background is no longer uniformly distributed, causing the false alarm rate to increase or the detection probability to decrease. A CFAR (constant false alarm) detection method based on clutter background statistical recognition.

Above-mentioned purpose of the invention is achieved through the following technical solutions:

Step 1. Start;

Step 2, input data RD spectrum;

Step 3, performing KL divergence partitioning on the RD spectrum in step 2 to obtain partitioned data;

Step 4, performing parameter estimation on the partitioned data to obtain estimated parameters;

Step 5, using the estimated parameters to normalize the background into an exponential distribution to obtain a normalized detection background;

Step 6: Perform CFAR detection on the normalized detection background to obtain a CFAR detection result.

Invention effect

The present invention combines KL divergence partition and detection background statistical estimation to propose a new detection method, which can segment the detection background, identify clutter and estimate parameters, and perform normalization processing, which can well control the complex clutter environment. probability of false alarm.

In this experiment, the distribution parameters obtained by processing the measured data are used to simulate the uniform area (the background clutter obeys the Weibull distribution), the clutter area (the background clutter obeys the lognormal distribution) and the clutter edge (the distribution characteristics of the reference units are different, and some subject to Weibull distribution, and the other part obeys lognormal distribution), by comparing the traditional CA (unit average), GO (maximum choice), SO (minimum choice), OS (ordered statistics)-CFAR, pair The Log-t under the normal distribution of numbers and the detection probability of CA, GO, SO, OS-CFAR estimated based on detection background statistics, the actual false alarm probability and the detection probability performance under the same actual false alarm probability are used to verify the algorithm.

The results are shown in Figures 1 to 9 and Tables 1 to 3. In the homogeneous area, the detection probability of CA, GO, SO, and OS-CFAR estimated based on detection background statistics is better than that of traditional CA, GO, SO, OS-CFAR, and Log-t under lognormal distribution. The results are shown in Figure 1 , the actual false alarm probability of Log-t under traditional CA, GO, SO, OS-CFAR, and lognormal distribution is lower than that of CA, GO, SO, OS-CFAR detectors estimated based on detection background statistics, but based on detection The detector based on background statistical estimation can keep the false alarm at about 0.01, and the results are shown in Table 1. In the case of interference targets, only OS-CFAR based on detection background statistical estimation and traditional OS-CFAR can maintain stable detection Performance, not affected by interference targets, the results are shown in Figure 2 to Figure 3; under the same actual false alarm probability, the detection probabilities of CA, GO, SO, and OS-CFAR estimated based on detection background statistics are comparable to those of traditional CA, The Log-t under GO, SO, OS-CFAR, and lognormal distribution are the same, and the results are shown in Figure 4. In the clutter area, although the detection probabilities of CA, GO, SO, and OS-CFAR estimated based on the detection background statistics are not improved compared with the Log-t under the traditional CA, GO, SO, OS-CFAR, and lognormal distribution , the results are shown in Figure 5, but the actual false alarm probability of traditional CA, GO, SO, and OS-CFAR is relatively high, and the CA, GO, SO, and OS-CFAR estimated based on the detection background statistics can keep the false alarm at the set value of about 0.01 , the results are shown in Table 2; in the presence of interfering targets, only OS-CFAR and traditional OS-CFAR based on detection background statistics can maintain stable detection performance and are not affected by interfering targets. The results are shown in Figures 6 to 7 Shown; under the same false alarm probability, the detection probability of CA, GO, SO, OS-CFAR estimated based on detection background statistics and the Log-t under the traditional CA, GO, SO, OS-CFAR, lognormal distribution Same, the result is shown in Figure 8. In the case of clutter edges, two cases are considered. One is that the unit to be detected is in the uniform area, and the other is that the detection unit is in the clutter area. When the detection unit is in the uniform area, the detection performance of the detector based on the statistical estimation of the detection background is improved. The results are shown in Figure 9. When the detection unit is in the clutter area, the statistical estimation based on the detection background can reduce the false alarm probability very well. The results are shown in Table 3. Through simulation analysis, when the actual detection background does not match the detector model, the performance of the detector will decline. Based on the background statistical estimation, the actual model is converted into the detector assumption model. The actual false alarm probability is equal to the theoretical false alarm probability, and the constant false alarm is maintained. It has good robustness. Therefore, CFAR detection based on statistical recognition of clutter background achieves the originally set target performance.

Table 1 Actual false alarm probability of various detectors in uniform area (set false alarm probability Pfa=0.01)

Table 2 The actual false alarm probability of various detectors in the clutter area (set false alarm probability Pfa=0.01)

Table 3 The actual false alarm probability of various detectors in the clutter edge area 2 (the detection unit is in the clutter area) (set the false alarm probability Pfa=0.01)

Description of drawings

Figure 1 is a performance comparison chart of CFAR detectors in a uniform area. In the figure, CA-CFAR is the unit average CFAR detection, GO-CFAR is the maximum selection CFAR detection, SO-CFAR is the minimum selection CFAR detection, OS-CFAR CFAR is an ordered statistical constant false alarm detection, and logt-CFAR is a Log-t detector under a lognormal distribution;

Figure 2 is a CFAR simulation performance comparison diagram in a uniform area with an interference target;

Figure 3 is a CFAR simulation performance comparison diagram in a uniform area with two interference targets;

Figure 4 is a performance comparison diagram of the uniform area CFAR detector under the same measured false alarm;

Figure 5 is a performance comparison diagram of the CFAR detector in the clutter area;

Figure 6 is a CFAR simulation performance comparison diagram in an interference target clutter area;

Figure 7 is a CFAR simulation performance comparison diagram in two interference target clutter areas;

Figure 8 is a performance comparison diagram of the CFAR detector in the clutter area under the same measured false alarm;

Fig. 9 is a performance diagram of various detectors in the clutter edge area 1 (the detection unit is in the uniform area);

Fig. 10 is a flowchart of the present invention;

Figure 11 is a brief flowchart of KL divergence partition;

Fig. 12 is a schematic diagram of a 5*5 square reference sliding window.

detailed description

Specific embodiment 1: This embodiment is described in conjunction with FIG. 10. A CFAR detection method based on clutter background statistical recognition in this embodiment is specifically prepared according to the following steps:

Step 1. Start;

Step 2, input data RD spectrum;

Step 3, carry out KL divergence (relative entropy) partition to the RD spectrum in step 2, obtain the data after partition;

Step 4, performing parameter estimation on the partitioned data to obtain estimated parameters;

Step 5, using the estimated parameters to normalize the background into an exponential distribution to obtain a normalized detection background;

Step 6: Perform CFAR detection on the normalized detection background to obtain a CFAR detection result.

Specific embodiment two: the difference between this embodiment and specific embodiment one is: the input data RD spectrum in the step two; the specific process is:

Load the data RD spectrum with matlab software.

Other steps and parameters are the same as those in Embodiment 1.

Specific embodiment three: the difference between this embodiment and specific embodiment one or two is: the RD spectrum in the step 2 is carried out KL divergence (relative entropy) partition in the described step 3, obtains the data after the partition; As shown in Figure 11 , the specific process is:

KL divergence is to distinguish the data of different probability distributions by calculating the difference between the two probability distributions P and Q;

For square-law detection, assuming that any range-Doppler unit obeys an exponential distribution, its PDF (probability density function) is

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

where: p(x) is the probability density function, β is the mean amplitude, x is the range-Doppler unit, and α is the center offset;

mean μ _x and variance Calculated by the following formula

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Based on the formulas (1), (2), and (3), the standard deviation σ _x of the measurement is used to identify the background distribution type (the PDF of different background distribution types is different); therefore, σ _x is used as a measure to classify the background; KL scatter The degree is defined as the measure of the dissimilarity between any two distributions, and this criterion is used to measure the dissimilarity of the distributions of two range-Doppler units; thus, the directional range definition for:

$\begin{array}{c}\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}\\ <span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\frac{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}}{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}\\ <span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; log</span>}\frac{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

where p ₁ (x) is the probability density function of the i-th range-Doppler unit, α ₁ is the center offset of the i-th range-Doppler unit, and β ₁ is the i-th range-Doppler The amplitude mean value of the Doppler unit, p ₂ (x) is the probability density function of the j-th Range-Doppler unit, α ₂ is the center offset of the j-th Range-Doppler unit, β ₂ is the j-th range-Doppler unit amplitude mean value, i, j are any positive integer;

Assuming that two distributions p ₁ (x) and p ₂ (x) have the same parameter α _i , the KL expression can be written as a function of the standard deviation

$\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

In the formula, I(p ₁ (x), p ₂ (x)) is a function of standard deviation, σ ₁ is the standard deviation of distribution p ₁ (x), σ ₂ is the standard deviation of distribution p ₂ (x);

In practice, it is very difficult to estimate the variance value of each unit. Therefore, considering that the distributions of all units are independent, the distance between the estimated variance of the entire background and the variance of all units within the reference window is used instead.

Step 31, as shown in Figure 11, take the square reference sliding window centered on each range-Doppler unit (P) in the RD spectrum data, and take N*N (N is an odd number, such as 5*5 or 9*9 The sliding window, the sliding window of 5*5 is taken in this patent), * is the multiplication sign; As shown in Figure 12;

Step 32, calculate the variance of the square reference sliding window (using matlab function std2 to calculate), replace the variance of the square reference sliding window center unit;

Step 33, calculating the variance of the entire RD spectral data;

Steps three and four, calculating the KL divergence of each range-Doppler unit according to steps three and two and three and three;

Step 35. When the KL divergence is smaller than the set threshold ξ, this area is called a clutter area and marked, and when the KL divergence is greater than the set threshold ξ, this area is called a uniform area and marked, and the The detection background is divided into uniform area and clutter area. Other steps and parameters are the same as those in Embodiment 1 or Embodiment 2.

Embodiment 4: This embodiment differs from Embodiment 1 to Embodiment 3 in that: in step 4, perform parameter estimation on the partitioned data to obtain estimated parameters; the specific process is:

The parameters of different regions are obtained by using the maximum likelihood estimation method.

Other steps and parameters are the same as those in Embodiments 1 to 3.

Specific embodiment five: the difference between this embodiment and one of the specific embodiments one to four is: in the step five, the estimated parameters are used to normalize the background and convert it into an exponential distribution to obtain the normalized detection background; the specific process for:

Let the cumulative distribution function of the random variable (this patent refers to Weibull distribution and Log-normal distribution) X be F, and the cumulative distribution function of the exponentially distributed random variable Y be G, then G ^-1 (F(X)) obeys the exponential distribution;

The cumulative distribution function of the Weibull distribution is as follows:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

In the formula, F ₁ (x) is the cumulative distribution function of Weibull distribution, B and C are the scale parameter and shape parameter of Weibull distribution respectively;

The cumulative distribution function of the Log-normal distribution is as follows:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; \&lsqb;</span>\frac{\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

Among them, F ₂ (x) is the cumulative distribution function of Log-normal distribution, erf is the error function, μ and σ are the scale parameter and shape parameter of Log-normal distribution, respectively;

The cumulative distribution function of the exponential distribution is as follows:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}}{{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

Among them, F ₃ (x) is the cumulative distribution function of the exponential distribution, and β ₃ is the mean parameter of the exponential distribution;

The Weibull distribution can be normalized and transformed into an exponential distribution through formula (6) and formula (8), and the derivation transformation formula is as follows:

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; exp</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; \&lsqb;</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; \&rsqb;</span>}^{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

Among them, x _exp is the amplitude value of the uniform area data unit in the radar echo distance-Doppler amplitude spectrum normalized by double parameters, x ₁ is the uniform area data in the detection background, B ₁ and C ₁ represent the uniform area data respectively The scale parameters and shape parameters estimated by the Weibull distribution of regional data take β ₃ =1 in actual conversion.

The Log-normal distribution can be normalized and transformed into an exponential distribution through formula (7) and formula (8), and the derivation transformation formula is as follows:

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; exp</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; lnx</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

Among them, x _exp1 is the amplitude value of the clutter area data unit in the RD spectrum after double-parameter normalization, x ₂ is the clutter area data unit in the detection background, μ ₂ and σ ₂ represent the logarithm positive of the clutter area respectively. The scale parameters and shape parameters estimated by the state distribution, take β ₃ =1 in the actual conversion;

The uniform area data and the clutter area data in the range-Doppler amplitude spectrum are transformed into exponential distribution by Weibull (Weibull) distribution and Log-normal (log-normal) distribution respectively.

Other steps and parameters are the same as in one of the specific embodiments 1 to 4.

Specific embodiment six: the difference between this embodiment and one of specific embodiments one to five is that: in the step six, CFAR detection is performed on the normalized detection background, and the CFAR detection result is obtained; the specific process is:

Using CA-CFAR (unit average constant false alarm detection), GO-CFAR (maximum selection constant false alarm detection), SO-CFAR (minimum selection constant false alarm detection), OS-CFAR (ordered statistical constant false alarm detection) Any one can get the CFAR test result.

Other steps and parameters are the same as one of the specific embodiments 1 to 5.

Adopt the following examples to verify the beneficial effects of the present invention:

Embodiment one:

In this embodiment, a CFAR detection method based on clutter background statistical recognition is specifically prepared according to the following steps:

In this experiment, the distribution parameters obtained by processing the measured data are used to simulate the uniform area (the background clutter obeys the Weibull distribution), the clutter area (the background clutter obeys the lognormal distribution) and the clutter edge (the distribution characteristics of the reference units are different, and some of them obey the Weibull distribution, the other part obeys the lognormal distribution) in these three cases, by comparing the Log-t under the traditional CA, GO, SO, OS-CFAR, lognormal distribution and the CA, GO based on the statistical estimation of the detection background , SO, OS-CFAR detection probability, the actual false alarm probability and the detection probability performance under the same actual false alarm probability to verify the algorithm.

The results are shown in Figures 1 to 9 and Tables 1 to 3. In the uniform area, the detection probability of CA, GO, SO, and OS-CFAR estimated based on the detection background statistics is better than the Log-t under the traditional CA, GO, SO, OS-CFAR, and lognormal distribution, and based on the detection background The statistically estimated detector can maintain the set false alarm, the results are shown in Figure 1 and Table 1; in the case of interference targets, only the OS-CFAR based on the statistical estimation of the detection background can maintain stable detection performance, and based on the detection background Statistically estimated SO-CFAR is second, and the results are shown in Figures 2 to 3; under the same false alarm probability, the detection probabilities of CA, GO, SO, and OS-CFAR estimated based on detection background statistics are comparable to traditional CA, GO, The Log-t under SO, OS-CFAR, and lognormal distribution are the same, and the results are shown in Figure 4. In the clutter area, although the detection probabilities of CA, GO, SO, and OS-CFAR estimated based on the detection background statistics are not improved compared with the Log-t under the traditional CA, GO, SO, OS-CFAR, and lognormal distribution , but traditional CA, GO, SO, and OS-CFAR have higher false alarm probability. CA, GO, SO, and OS-CFAR estimated based on detection background statistics can keep the false alarm at the set value. The results are shown in Figure 5 and Table 2 As shown; in the case of interfering targets, only OS-CFAR based on detection background statistics can maintain stable detection performance, and SO-CFAR based on detection background statistics is second. The results are shown in Figures 6 to 7; the same false alarm Under probability, the detection probabilities of CA, GO, SO, and OS-CFAR estimated based on detection background statistics are the same as the Log-t under traditional CA, GO, SO, OS-CFAR, and lognormal distribution. The results are shown in Figure 8 Show. In the case of clutter edges, two cases are considered. One is that the unit to be detected is in the uniform area, and the other is that the detection unit is in the clutter area. When the detection unit is in the uniform area, the detection performance of the detector based on the statistical estimation of the detection background is improved. In the clutter area, statistical estimation based on the detection background can reduce the false alarm probability very well, and the results are shown in Figure 9 and Table 3. Through simulation analysis, when the actual detection background does not match the detector model, the performance of the detector will decline. Based on the background statistical estimation, the actual model is converted into the detector assumption model. The actual false alarm probability is equal to the theoretical false alarm probability, and the constant false alarm is maintained. It has good robustness. Therefore, CFAR detection based on statistical recognition of clutter background achieves the originally set target performance.

Table 1 Actual false alarm probability of various detectors in uniform area (set false alarm probability Pfa=0.01)

Table 2 The actual false alarm probability of various detectors in the clutter area (set false alarm probability Pfa=0.01)

Table 3 The actual false alarm probability of various detectors in the clutter edge area 2 (the detection unit is in the clutter area) (set the false alarm probability Pfa=0.01)

The present invention can also have other various embodiments, without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and deformations according to the present invention, but these corresponding changes and deformations are all Should belong to the scope of protection of the appended claims of the present invention.

Claims (6)

1., based on the CFAR detection method that clutter background statistics identifies, it is characterized in that what a kind of CFAR detection method identified based on clutter background statistics was specifically carried out according to following steps:

Step one, beginning;

Step 2, input data RD spectrum;

Step 3, in step 2 RD spectrum carry out KL divergence subregion, obtain the data after subregion;

Step 4, parameter estimation is carried out to the data after subregion, obtain the parameter estimated;

The parameter that step 5, utilization are estimated converts background normalization to exponential distribution, obtains the detection background after normalization;

Step 6, CFAR detection is carried out to the detection background after normalization, draw CFAR testing result; CFAR is CFAR.

2. according to claim 1 a kind of based on clutter background statistics identify CFAR detection method, it is characterized in that: input in described step 2 data RD spectrum; Detailed process is:

Data RD spectrum is loaded into matlab software.

3. according to claim 2 a kind of based on clutter background statistics identify CFAR detection method, it is characterized in that: in described step 3 in step 2 RD spectrum carry out KL divergence subregion, obtain the data after subregion; Detailed process is:

KL divergence is by calculating two differences between probability distribution P and Q, realizing the differentiation to different probability distributed data;

To quadratic detection, suppose any one distance-Doppler unit obeys index distribution, its PDF is

$p\left(x\right)=\frac{1}{\mathrm{\&beta;}}{e}^{-(x-\mathrm{\&alpha;})/\mathrm{\&beta;}}---\left(1\right)$

In formula: p (x) is probability density function, and β is amplitude equalizing value, x is distance-Doppler unit, and α is the center offset of distance-Doppler unit;

Average μ _xand variance calculated by formula below

${\mathrm{\&mu;}}_x={\&Integral;}_{\mathrm{\&alpha;}}^{\mathrm{\&infin;}}x\&CenterDot;p\left(x\right)dx=\mathrm{\&alpha;}+\mathrm{\&beta;}---\left(2\right)$

${\mathrm{\&sigma;}}_x^2={\&Integral;}_{\mathrm{\&alpha;}}^{\mathrm{\&infin;}}{x}^2\&CenterDot;p\left(x\right)dx-{\mathrm{\&mu;}}_x^2={\mathrm{\&beta;}}^2---\left(3\right)$

Based on formula (1), (2), (3), utilize the standard deviation sigma measured _xdistinguish background distributions type; Therefore σ is utilized _xbackground of classifying is estimated as one; K-L divergence is defined as any two kinds of distribution diversities and measures, and adopts this criterion to weigh the distributional difference of two distance-Doppler unit; Therefore, the orientation distance of same two separate units with reference to the obedience different distributions in sliding window is defined as:

$\begin{array}{c}I\left(p_1\right(x),p_2\left(x\right))={\&Integral;}_{{\mathrm{\&alpha;}}_1}^{\mathrm{\&infin;}}{p}_1\left(x\right)log\left(\frac{p_1\left(x\right)}{p_2\left(x\right)}\right)dx\\ ={\&Integral;}_{{\mathrm{\&alpha;}}_1}^{\mathrm{\&infin;}}\frac{1}{{\mathrm{\&beta;}}_1}{e}^{-(x-{\mathrm{\&alpha;}}_1)/{\mathrm{\&beta;}}_1}log\left(\frac{\frac{1}{{\mathrm{\&beta;}}_1}{e}^{-(x-{\mathrm{\&alpha;}}_1)/{\mathrm{\&beta;}}_1}}{\frac{1}{{\mathrm{\&beta;}}_2}{e}^{-(x-{\mathrm{\&alpha;}}_2)/{\mathrm{\&beta;}}_2}}\right)dx\\ =\log \frac{{\mathrm{\&beta;}}_2}{{\mathrm{\&beta;}}_1}+\frac{{\mathrm{\&alpha;}}_1+{\mathrm{\&beta;}}_1-{\mathrm{\&alpha;}}_2-{\mathrm{\&beta;}}_2}{{\mathrm{\&beta;}}_2}\end{array}---\left(4\right)$

In formula, p ₁x () is the probability density function of i-th distance-Doppler unit, α ₁the center offset of i-th distance-Doppler unit, β ₁the amplitude equalizing value of i-th distance-Doppler unit, p ₂x () is the probability density function of a jth distance-Doppler unit, α ₂the center offset of a jth distance-Doppler unit, β ₂be the amplitude equalizing value of a jth distance-Doppler unit, i, j value is any positive integer;

Suppose two distribution p ₁(x) and p ₂x () has identical parameters α _i, KL expression formula can be write as the function of standard deviation

$I\left(p_1\right(x),p_2(x\left)\right)=log\frac{{\mathrm{\&sigma;}}_2}{{\mathrm{\&sigma;}}_1}+\frac{{\mathrm{\&sigma;}}_1-{\mathrm{\&sigma;}}_2}{{\mathrm{\&sigma;}}_2}---\left(5\right)$

In formula, I (p ₁(x), p ₂(x)) be the function of standard deviation, σ ₁it is distribution p ₁the standard deviation of (x), σ ₂it is distribution p ₂the standard deviation of (x);

Step 3 one, the square reference of getting in RD modal data centered by each distance-Doppler unit slide window, and get N*N, * is multiplication sign;

Step 3 two, calculating square, with reference to the variance of sliding window, replace the square variance with reference to sliding window center unit;

Step 3 three, calculate the variance of whole RD modal data;

Step 3 four, calculate the KL divergence of each distance-Doppler unit according to step 3 two and step 3 three;

Step 3 five, when KL divergence is less than the threshold xi of setting, this region is called clutter region and marks, when KL divergence is greater than the threshold xi of setting, this region be called homogeneous area and mark, detection background being divided into homogeneous area and clutter region.

4. according to claim 3 a kind of based on clutter background statistics identify CFAR detection method, it is characterized in that: in described step 4, parameter estimation is carried out to the data after subregion, obtain estimate parameter; Detailed process is:

Maximum Likelihood is utilized to obtain the parameter of zones of different.

5. according to claim 4 a kind of based on clutter background statistics identify CFAR detection method, it is characterized in that: utilize the parameter of estimation to convert background normalization to exponential distribution in described step 5, obtain the detection background after normalization; Detailed process is:

If the cumulative distribution function of stochastic variable X is F, the cumulative distribution function of exponential random variable Y is G, then G ^-1(F (X)) obeys index distribution;

The cumulative distribution function of Weibull distribution is as follows:

$F_1\left(x\right)=1-\exp \&lsqb;-{\left(\frac{x}{B}\right)}^C\&rsqb;---\left(6\right)$

In formula, F ₁x () is the cumulative distribution function of Weibull distribution, B and C is scale parameter and the form parameter of Weibull distribution respectively;

The cumulative distribution function of Log-normal distribution is as follows:

$F_2\left(x\right)=\frac{1}{2}+\frac{1}{2}erf\&lsqb;\frac{ln\left(x\right)-\mathrm{\&mu;}}{\mathrm{\&sigma;}\sqrt{2}}\&rsqb;---\left(7\right)$

Wherein, F ₂x () is the cumulative distribution function of Log-normal distribution, erf is error function, μ and σ is scale parameter and the form parameter of Log-normal distribution respectively;

The cumulative distribution function of exponential distribution is as follows:

$F_3\left(x\right)=1-\exp \&lsqb;-\frac{x}{{\mathrm{\&beta;}}_3}\&rsqb;---\left(8\right)$

Wherein, F ₃x () is the cumulative distribution function of exponential distribution, β ₃for the Mean Parameters of exponential distribution;

Be converted to exponential distribution by the normalization that to be distributed by Weibull of formula (6) and formula (8), derivation transformation for mula is as follows:

$x_{\exp }={\mathrm{\&beta;}}_3{\&lsqb;\frac{x_1}{B_1}\&rsqb;}^{C_1}---\left(9\right)$

Wherein, x _expfor the range value of homogeneous area data cell after two-parameter normalization in radar return distance-Doppler amplitude spectrum, x ₁for homogeneous area data in detection background, B ₁and C ₁represent the scale parameter that homogeneous area data Weibull distribution is estimated and form parameter respectively, in actual converted, get β ₃=1;

Be converted into exponential distribution by the normalization that to be distributed by Log-normal of formula (7) and formula (8), derivation transformation for mula is as follows:

$x_{\exp 1}=-{\mathrm{\&beta;}}_{3}ln(\frac{1}{2}-\frac{1}{2}erf(\frac{{\mathrm{lnx}}_2-{\mathrm{\&mu;}}_2}{\sqrt{2}{\mathrm{\&sigma;}}_2}\left)\right)---\left(10\right)$

Wherein, x _exp1for the range value of clutter area data unit after two-parameter normalization in RD spectrum, x ₂for the clutter area data unit in detection background, μ ₂and σ ₂represent the scale parameter that clutter region lognormal distribution is estimated and form parameter respectively, in actual converted, get β ₃=1.

6. according to claim 5 a kind of based on clutter background statistics identify CFAR detection method, it is characterized in that: in described step 6, CFAR detection is carried out to the detection background after normalization, draw CFAR testing result; Detailed process is:

In the detection of employing CA-CFAR, MAXIMUM SELECTION CFAR detection, minimum selection CFAR detection, Ordered Statistic CFAR detection, any one can draw CFAR testing result.