Clutter map detection method based on OTSU-CCA

Abstract

The invention relates to a clutter map detection method based on OTSU-CCA, and belongs to the technical field of radar target detection. Comprising the following steps: 1) performing KS normality test on a continuous scanning period sample of a to-be-detected unit, and verifying whether the continuous scanning period sample of the to-be-detected unit obeys exponential distribution or not; 2) sequencing the samples in the continuous scanning period of the unit to be detected according to a sequence from small to large; 3) traversing each sorted sample, and calculating the corresponding between-class variance of the sample; the sorted sample sequence number corresponding to the maximum between-class variance value is the demarcation point of the sample containing the interference target and the sample not containing the interference target; and 4) carrying out detection processing by adopting a CCA-CFAR detector, and further carrying out screening. According to the method, the adaptive screening process is added, so that interference target samples in continuous scanning period samples of the azimuth-distance unit to be detected can be eliminated, robust estimation of the background power level is realized, and the detection performance is improved.

Description

Technical Field

[0001] The invention relates to a clutter map detection method based on OTSU-CCA, and belongs to the technical field of radar target detection. Background Art

[0002] When clutter changes dramatically in the spatial domain, the traditional spatial constant false alarm rate (CFAR) algorithm will increase the CFAR loss due to the small number of available reference unit samples. In this case, if the clutter changes smoothly in the time domain, the time domain CFAR algorithm, namely the clutter map (CM) algorithm, can be used. The clutter map algorithm was first proposed by Nitzberg et al. in 1986. By setting an independent threshold for each azimuth-range unit, it can effectively solve the problem of inaccurate background power level estimation caused by uneven background in the spatial domain CFAR algorithm.

[0003] According to the different selected spatial ranges, clutter map technology can be divided into clutter map surface technology and clutter map point technology, or according to different dimensions, it can be divided into plane clutter map and three-dimensional clutter map. Its background estimation is usually generated by recursively updating the background power level of the average estimated unit or the previous clutter storage value. Recursive updating weakens the impact of echoes that are far apart in time on the background estimation, and the background estimation is obtained by weighting the different trust levels selected for the current echo and the background estimation at previous times. However, the recursive update method cannot overcome the impact of interference targets, especially low-speed targets, on the background estimation. When the interference target stays in the azimuth-range unit for a long time, the samples obtained by multiple scans all contain the interference target, resulting in increasingly serious deviations in the background estimation, which in turn deteriorates the detection performance of the clutter map. Summary of the invention

[0004] In order to solve the shortcomings of the above-mentioned prior art, a clutter map detection method based on OTSU-CCA is provided, which is denoted as OTSU-CCA-CM detector. By adding an adaptive screening process, the method can eliminate the interference target samples in the continuous scanning cycle samples of the azimuth-range unit to be detected, thereby achieving a robust estimation of the background power level and improving the detection performance.

[0005] The clutter map detection method based on OTSU-CCA of the present invention is special in that it comprises the following steps:

[0006] Step 1: Perform a KS normality test on the continuous scanning cycle samples of the unit to be detected to verify whether the continuous scanning cycle samples of the unit to be detected obey the exponential distribution; if the verification result shows that the continuous scanning cycle samples of the unit to be detected obey the exponential distribution, directly enter the CCA-CFAR detection process in step 4; if the verification result shows that the continuous scanning cycle samples of the unit to be detected do not obey the exponential distribution, execute the maximum inter-class difference algorithm, recorded as the OTSU algorithm, that is, enter step 2-step 3;

[0007] Step 2: sort the samples in the continuous scanning cycle of the unit to be detected in ascending order;

[0008] Step 3: Traverse each sorted sample and calculate its corresponding inter-class variance; the sorted sample number corresponding to the maximum inter-class variance is the dividing point between samples with interference targets and samples without interference targets;

[0009] Step 4: Use CCA-CFAR detector to perform detection processing and further screening.

[0010] Preferably, the specific algorithm of the CCA-CFAR detector is:

[0011] The process of removing the spike signal in the reference unit sample by the CCA-CFAR detector is as follows:

[0012] Assume x1,x2,···,x R is the R reference unit sample value, T0 is the number of reference units R and the given virtual

[0013] Alarm probability P fa (0) The corresponding threshold factor, find the first sum value, that is:

[0014]

[0015] Each reference cell sample is then compared to the threshold S1 as follows:

[0016]

[0017] The reference unit samples that exceed this threshold are removed from the sum value, and the remaining reference unit samples are used to form a new sum value:

[0018]

[0019] Among them, j1 is the number of reference unit samples that are eliminated, and k1 is the subscript of the remaining reference unit samples;

[0020] The threshold factor T1 is also introduced, which corresponds to the number of reference units R1 and the false alarm probability P fa(1) , and then the remaining

[0021] The remaining reference unit samples are compared with the following threshold S2:

[0022]

[0023] The reference unit sample values ​​exceeding this threshold are eliminated, j2 is the number of reference unit samples eliminated in this operation, and the remaining R2=R0-j1-j2 reference unit samples are used to form a new sum value:

[0024]

[0025] The above process continues until no samples exceeding the threshold are detected in the reference unit samples;

[0026] Probability P fa (i) is the false alarm probability of each screening layer, and the total average number of false alarms is Therefore, the total false alarm probability P fa for:

[0027]

[0028] The simplest way to keep the false alarm rate constant is to make all P fa (i) Equal to the predetermined P fa value.

[0029] Preferably, the specific process of the OTSU algorithm is:

[0030] 1) Sort the samples in the continuous scanning cycle of the unit to be detected by amplitude, expressed as:

[0031] x (1) ≤x (2) ≤…≤x (n) (7)

[0032] Among them, x (i) , (i=1,…,n) represents the sorted continuous scanning cycle samples of the units to be detected, and n is the number of continuous scanning cycles;

[0033] 2) Assume that there is a dividing point k to divide the above samples into two categories, denoted as C1, which represents the samples without interference targets in the continuous scanning cycle of the unit to be detected, and C2, which represents the samples with interference targets in the continuous scanning cycle of the unit to be detected; calculate the cumulative means of these two categories of samples, denoted as m1(k) and m2(k), and denote the probabilities of samples being divided into C1 and C2 as p1(k) and p2(k), respectively, so:

[0034] p1(k)+p2(k)=1 (8)

[0035] in,

[0036] The calculation formula for the inter-class variance corresponding to the cutoff point position k is:

[0037] σ 2 (k)=p1(k)*p2(k)*(m1(k)-m2(k)) 2 (9)

[0038] Search k from 1 to n so that the above formula σ 2 (k) When k reaches the maximum, it is the dividing point between C1 and C2, denoted as k0;

[0039] 3) From the above steps, we can see that x (k0+1) ,…,x (n) That is, the interference target sample in the continuous scanning cycle of the unit to be detected, That is, the samples in the continuous scanning cycle that have been screened in the unit to be detected and can be preliminarily used for clutter background estimation.

[0040] The clutter map detection method based on OTSU-CCA of the present invention has the following beneficial effects:

[0041] 1. For the scenario where there is no interference target in the continuous scanning cycle of the unit to be detected, the detection performance of the OTSU-CCA-CM detector is very close to that of the clutter map CFAR detector, denoted as CM-CFAR detector, which shows that the OTSU-CCA-CM detector proposed in this paper is suitable for the scenario where there is no interference target.

[0042] 2. For both the cases where there are few interfering target samples and many interfering target samples in the continuous scanning cycle of the unit to be detected, the performance of the CM-CFAR detector drops sharply, while the detection performance of the OTSU-CCA-CM detector is only slightly reduced compared to the case without interfering targets; this shows that the proposed OTSU-CCA-CM detector can effectively solve the target detection problem when there are interfering target samples in the continuous scanning cycle of the unit to be detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 :It is the block diagram of CCA-CFAR detector;

[0044] Figure 2 : is the sample and deletion threshold diagram in the continuous scanning cycle of the unit to be detected in the case of no interference target;

[0045] Figure 3 : Samples and deletion threshold diagram in the continuous scanning cycle of the unit to be detected under the condition of less interference target;

[0046] Figure 4 : It is the sample and deletion threshold diagram in the continuous scanning cycle of the unit to be detected under the situation of multiple interference targets;

[0047] Figure 5 : In the case of multiple interference targets, the samples in the continuous scanning cycle of the unit to be detected are arranged in order from small to large;

[0048] Figure 6 : It is the inter-class variance diagram under the condition of multiple interference targets;

[0049] Figure 7 :Performance curves for comparing two CFAR detectors (CM, OTSU-CCA-CM) under three interference target situations;

[0050] Figure 8 : Continuous scanning period samples and deletion threshold maps obtained by two methods;

[0051] Fig. 9 : It is the between-class variance and the cut-off point diagram. DETAILED DESCRIPTION

[0052] Example 1

[0053] This embodiment describes in detail a clutter map detection method based on OTSU-CCA of the present invention. The processing flow of the present invention is described in detail below:

[0054] 1) Perform a KS (Kolmogorov-Smirnov) normality test on the samples of the continuous scanning cycle of the unit to be detected to verify whether the samples of the continuous scanning cycle of the unit to be detected obey the exponential distribution. If the verification result shows that the samples of the continuous scanning cycle of the unit to be detected obey the exponential distribution, directly enter the CCA (Censored Cell-Averaging, delete the unit average)-CFAR detection process in step 4); if the verification result shows that the samples of the continuous scanning cycle of the unit to be detected do not obey the exponential distribution, execute the maximum inter-class difference algorithm, recorded as the OTSU algorithm, that is, enter steps 2)-3);

[0055] 2) The samples in the continuous scanning cycle of the unit to be detected are sorted in order from small to large.

[0056] 3) Traverse each sorted sample and calculate its corresponding inter-class variance OTSU. The sorted sample number corresponding to the maximum inter-class variance is the dividing point between samples with interference targets and samples without interference targets.

[0057] 4) CCA-CFAR detection was used for further screening.

[0058] The CCA algorithm and the OTSU algorithm are specifically:

[0059] 1) CCA-CFAR detector (Deleted Cell Average CFAR detector)

[0060] The basic process of the CCA-CFAR detector to remove possible spike signals in the reference unit samples (ie, the reference unit samples may contain samples of interfering targets or other samples inconsistent with the statistical characteristics of the detection unit background) is as follows.

[0061] Assume x1,x2,···,x R is the R reference unit sample value, T0 is the number of reference units R and the given false alarm probability P fa (0) The corresponding threshold factor, find the first sum value, that is:

[0062]

[0063] Each reference cell sample is then compared to the following threshold:

[0064]

[0065] The reference unit samples that exceed this threshold are removed from the sum value, and the remaining reference unit samples are used to form a new sum value:

[0066]

[0067] Among them, j1 is the number of reference unit samples that are eliminated, and k1 is the subscript of the remaining reference unit samples.

[0068] The threshold factor T1 is also introduced, which corresponds to the number of reference units R1 and the false alarm probability P fa (1) The remaining reference cell samples are then compared to the following threshold:

[0069]

[0070] Remove the reference unit sample values ​​that exceed this threshold, j2 is the number of reference unit samples removed in this operation. Use the remaining R2 = R0-j1-j2 reference unit samples to form a new sum value:

[0071]

[0072] The above process continues until no samples exceeding the threshold are detected in the reference unit samples. Obviously, this algorithm always converges. Therefore, after several operations, the absence of samples exceeding the threshold in the reference unit samples is the termination criterion of the CCA-CFAR detection method.

[0073] Probability P fa (i) is the false alarm probability of each screening layer. The total average number of false alarms is Therefore, the total false alarm probability is:

[0074]

[0075] The simplest way to keep the false alarm rate constant is to make all P fa (i) Equal to the predetermined P fa value.

[0076] Figure 1 The principle block diagram of the CCA-CFAR detection process is given. The spike signal removal at each level includes three steps: ① summing the spike power; ② spike signal counting; ③ zero value insertion (inserting into the buffer to replace the spike signal power). When all reference unit samples in the buffer pass the comparator, the first cycle is completed. If the output of the spike signal counter is not zero, the sum of the spike signals is subtracted from the sum of the reference unit sample values ​​at the beginning of the next cycle, and then the method of the first cycle is repeated. The only difference is that the threshold factor of this cycle is determined by the number of reference units remaining after the spikes are removed. If no spike signal passes the threshold in the current cycle, the removal process is over. Therefore, the detection threshold formed by the sum of the reference unit samples obtained in the last cycle and the threshold factor of this cycle is used to make a threshold decision on the detection unit samples, thereby completing the CCA-CFAR detection process.

[0077] 2) Maximum Inter-Class Difference Method (OTSU)

[0078] OTSU, also known as the maximum inter-class difference method, was originally used to determine the threshold for image segmentation. It performs binary segmentation of foreground and background images by calculating the inter-class variance. This embodiment draws on the idea of ​​this method to sort target samples, thereby solving the problem of "the number of interfering targets accounts for a large proportion of the number of reference samples", and provides a prerequisite for using the CCA algorithm to delete only the remaining interfering target samples. The specific process is:

[0079] 1) Sort the samples in the continuous scanning cycle of the unit to be detected by amplitude, expressed as:

[0080] x (1) ≤x (2) ≤…≤x (n) (7)

[0081] Among them, x (i) , (i=1,…,n) represents the sorted continuous scanning cycle samples of the units to be detected, and n is the number of continuous scanning cycles;

[0082] 2) Assume that there is a dividing point k to divide the above samples into two categories, denoted as C1, which represents the samples without interference targets in the continuous scanning cycle of the unit to be detected, and C2, which represents the samples with interference targets in the continuous scanning cycle of the unit to be detected; calculate the cumulative means of these two categories of samples, denoted as m1(k) and m2(k), and denote the probabilities of samples being divided into C1 and C2 as p1(k) and p2(k), respectively, so:

[0083] p1(k)+p2(k)=1 (8)

[0084] in,

[0085] The calculation formula for the inter-class variance corresponding to the cutoff point position k is:

[0086] σ 2 (k)=p1(k)*p2(k)*(m1(k)-m2(k)) 2 (9)

[0087] Search k from 1 to n so that the above formula σ 2 (k) When k reaches the maximum, it is the dividing point between C1 and C2, denoted as k0;

[0088] 3) From the above steps, we can see that x (k0+1) ,…,x (n) That is, the interference target sample in the continuous scanning cycle of the unit to be detected, That is, the samples in the continuous scanning cycle that have been screened in the unit to be detected and can be preliminarily used for clutter background estimation.

[0089] Example 2

[0090] This embodiment designs three interference target situations. The first is no interference target, that is, in the continuous scanning cycle, there is no interference target sample in the azimuth-range resolution unit to be detected, such as Figure 2 As shown; the second is to have fewer interference target samples. In the 8th to 13th scanning cycles, a total of 6 interference target samples are set, and their intensity is 18dB, as shown Figure 3 As shown; the third is multiple interference target samples, in the 8th to 13th, 24th to 29th, 30th to 35th, 51st to 56th scanning cycles, a total of 24 interference target samples are set, and the intensity is 18dB, as shown Figure 4 shown.

[0091] Under different numbers of interference target samples, the background mean is estimated by using the unit average to obtain the deletion threshold of the first step. Figure 2 , Figure 3 , Figure 4 As shown by the black solid line in Figure 4It can be seen that under the condition of multiple interfering targets, the deletion threshold is greatly raised, higher than the samples in the continuous scanning cycle of all units to be detected, so the interfering targets cannot be effectively removed. This shows that the multi-step deletion strategy directly using the CCA algorithm cannot effectively remove the interfering target samples in the case of multiple interfering targets.

[0092] KS tests are performed on samples in the continuous scanning cycle of the unit to be detected under three interference target situations to verify whether they conform to the exponential distribution. In the case of multiple interference targets, the samples in the continuous scanning cycle of the unit to be detected are considered not to conform to the exponential distribution after testing; while the samples in the case of no interference targets and the case of few interference targets are considered to conform to the exponential distribution. It should be noted that in the case of few interference targets, since the number of interference target samples in the continuous scanning cycle is small and their intensity is limited, the influence of the interference target energy on the overall energy of the sample is limited, and it does not change the law of its exponential distribution. However, the interference targets exist objectively. Therefore, after implementing the KS test, the algorithm in this paper also sends the samples that conform to the exponential distribution into the CCA algorithm processing step to eliminate the possible samples with few interference targets.

[0093] When the samples in the continuous scanning cycle of the unit to be detected are considered to be inconsistent with the exponential distribution, the OTSU algorithm is executed to solve the demarcation point between the interference target samples and the non-interference target samples, which is the core step of this method. First, the samples in the continuous scanning cycle of the unit to be detected are sorted in order from small to large, such as Figure 5 shown.

[0094] Then, traverse each sorted sample and calculate its corresponding inter-class variance, such as Figure 6 As shown in the figure. The sorted sample number corresponding to the maximum inter-class variance is the dividing point between samples with interference targets and samples without interference targets, i.e., k0 in step 3). According to the simulation conditions set, the interference target should start to appear at the 40th position after sorting, i.e., the true value of k0 is 40. The estimated value of k0 obtained in step 3) is consistent with its true value, indicating that the OTSU algorithm can effectively eliminate samples with interference targets.

[0095] Example 3

[0096] This embodiment is directed to a set of measured data measured in a certain distance-azimuth unit of a certain shore-based sea surveillance radar. Figure 8 and Figure 8 The processing results are given. Figure 8In the figure, the black dotted line represents a continuous scanning cycle sample in the range-azimuth unit; the black dashed line represents the deletion threshold of the reference sample obtained only by CCA. Obviously, due to the large proportion of interfering target samples, the multi-step deletion strategy of CCA does not delete the interfering samples; the black solid line represents the deletion threshold of the reference sample obtained by the method proposed in this paper. Obviously, the effective deletion of interfering target samples is achieved. Fig. 9 The inter-class variance and the dividing point between interference target samples and background samples obtained by the OTSU method after the samples are sorted in order in the continuous scanning period are given.

Claims

1. A clutter map detection method based on OTSU-CCA, characterized in that The following steps are involved: Step 1: Perform a KS normality test on the continuous scanning cycle samples of the unit to be detected to verify whether the continuous scanning cycle samples of the unit to be detected obey the exponential distribution; if the verification result shows that the continuous scanning cycle samples of the unit to be detected obey the exponential distribution, directly enter the CCA-CFAR detection process in step 4; if the verification result shows that the continuous scanning cycle samples of the unit to be detected do not obey the exponential distribution, execute the maximum inter-class difference algorithm, recorded as the OTSU algorithm, that is, enter step 2-step 3; Step 2: sort the samples in the continuous scanning cycle of the unit to be detected in ascending order; Step 3: Traverse each sorted sample and calculate its corresponding inter-class variance; the sorted sample number corresponding to the maximum inter-class variance is the dividing point between samples with interference targets and samples without interference targets; Step 4: Use CCA-CFAR detector to perform detection processing and further screening.

2. A clutter map detection method based on OTSU-CCA according to claim 1, characterized in that The specific algorithm of the CCA-CFAR detector is: The process of removing the spike signal in the reference unit sample by the CCA-CFAR detector is as follows: Assume x1,x2,···,x R is the R reference unit sample value, T0 is the number of reference units R and the given false alarm probability P fa (0) The corresponding threshold factor, find the first sum value, that is: Each reference cell sample is then compared to the threshold S1 as follows: The reference unit samples that exceed this threshold are removed from the sum value, and the remaining reference unit samples are used to form a new sum value: Among them, j1 is the number of reference unit samples that are eliminated, and k1 is the subscript of the remaining reference unit samples; The threshold factor T1 is also introduced, which corresponds to the number of reference units R1 and the false alarm probability P fa (1) , and then compare the remaining reference unit samples with the following threshold S2: The reference unit sample values ​​exceeding this threshold are eliminated, j2 is the number of reference unit samples eliminated in this operation, and the remaining R2=R0-j1-j2 reference unit samples are used to form a new sum value: The above process continues until no samples exceeding the threshold are detected in the reference unit samples; the probability P fa (i) is the false alarm probability of each screening layer, and the total average number of false alarms is Therefore, the total false alarm probability P fa for: The simplest way to keep the false alarm rate constant is to make all P fa (i) Equal to the predetermined P fa value.

3. A clutter map detection method based on OTSU-CCA according to claim 1 or 2, characterized in that The specific process of the OTSU algorithm is as follows: 1) Sort the samples in the continuous scanning cycle of the unit to be detected by amplitude, expressed as: x (1) ≤x (2) ≤…≤x (n) (7) Among them, x (i) , (i=1,…,n) represents the sorted continuous scanning cycle samples of the units to be detected, and n is the number of continuous scanning cycles; 2) Assume that there is a dividing point k to divide the above samples into two categories, denoted as C1, which represents the samples without interference targets in the continuous scanning cycle of the unit to be detected, and C2, which represents the samples with interference targets in the continuous scanning cycle of the unit to be detected; calculate the cumulative means of these two categories of samples, denoted as m1(k) and m2(k), and denote the probabilities of samples being divided into C1 and C2 as p1(k) and p2(k), respectively, so: p1(k)+p2(k)=1 (8) in, The calculation formula for the inter-class variance corresponding to the cutoff point position k is: σ 2 (k)=p1(k)*p2(k)*(m1(k)-m2(k)) 2 (9) Search k from 1 to n so that the above formula σ 2 (k) When k reaches the maximum, it is the dividing point between C1 and C2, denoted as k0; 3) From the above steps, we can see that That is, the interference target sample in the continuous scanning cycle of the unit to be detected, That is, the samples in the continuous scanning cycle that have been screened in the unit to be detected and can be preliminarily used for clutter background estimation.

Images