A refined radar detection method based on clutter characteristic perception

Abstract

The object of the present invention is to provide a method for fine target detection in the presence of clutter and noise. In the case where both clutter background and noise background coexist, the first problem to be solved is the segmentation of the clutter area and the noise area. The present invention utilizes the differences in statistical characteristics between clutter signals and noise signals to segment the clutter area and the noise area. Next, target detectors are designed respectively in the clutter area and the noise area to achieve fine target detection and optimize the detection performance of the radar system.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar signal processing, and particularly relates to a target detection method. Background Art

[0002] In the data collected by a radar, in addition to the signals reflected by targets such as airplanes and ships, a large number of clutter signals and noise signals are often included. It is relatively easy to detect target signals in noise (this type of problem is called target detection in a noise background), and the technology is already relatively mature; on the contrary, it is not easy to solve the detection of target signals in clutter (this type of problem is called target detection in a clutter background). To effectively solve the problem of target detection in a clutter background, it is necessary to accurately and reliably perform clutter-noise segmentation to identify whether the detection background is a noise background or a clutter background. Clutter is the echo signal caused by the ocean surface or the land surface. Compared with the echo signals of targets such as airplanes and ships, the number of clutter is huge and appears disorderly, but some statistical characteristics of clutter are regular. The present invention is to perform clutter-noise segmentation by using the statistical characteristics of radar signals and perform refined target detection according to the segmentation results.

[0003] Identifying whether the detection background is a noise background or a clutter background by performing clutter-noise segmentation has a wide range of applications in engineering. For example, the literature [Shang Lina et al., Partition processing scheme based on airborne radar clutter distribution, Modern Radar, 2009, Vol. 31, No. 10] proposed that after conventional PD (Pulsed-Doppler) processing, subsequent processing is carried out according to the classification of clutter background and noise background: STAP (Space-Time Adaptive Processing) processing is adopted in the clutter background, and a frequency-domain coherent accumulation method is adopted in the noise background.

[0004] The literature [Li Jingjing et al., Research on clutter partition suppression method for airborne fire control radar, Electronic Science and Technology, 2013, Vol. 26, No. 4] improved the method of the foregoing literature. A clutter suppression method is adopted in the clutter background area, and a constant false alarm rate detection method is directly adopted in the noise area. In addition, this literature proposed a processing flow based on clutter-noise segmentation: in the range-frequency two-dimensional data, the distribution of the power of the echo with frequency is statistically analyzed, and the inflection point of its change is judged as the boundary point of the clutter. The disadvantages of this method are: First, large-sized targets, such as large ships on the sea surface, will affect the position of the inflection point of the echo power intensity, which may cause misjudgment; Second, if the clutter only exists in a certain range, such as only in the near range and not in the far range, then statistically analyzing the change of clutter power with frequency according to the method of this literature is very inaccurate, and it also exposes that the method of this literature cannot distinguish whether there is clutter at a certain range. The method of this literature can only distinguish whether there is clutter at a certain frequency and cannot achieve refined processing.

[0005] The literature [Zhou Ming et al., Research on Land-Sea Separation and Sea Area Partitioning Algorithms for Sea Surface Target Detection, Journal of Radars, 2019, Vol. 8, No. 3] proposed a fast and accurate land-sea separation method, which utilized the phase correlation of the echo sequence as the discrimination basis. Secondly, for the regions judged as sea clutter, partitioning was carried out based on the grazing angle, and the sea clutter model was estimated on each partition, and a suitable detector was selected according to the sea clutter model. The land-sea separation algorithm in this literature performed a traversal process on all data. If the clutter region could be quickly and accurately determined before land-sea separation, and then further distinguished whether it was ground clutter or sea clutter for the clutter region, the processing process would be more efficient and reasonable. In the sea clutter region, this literature partitioned according to the grazing angle without considering how to demarcate the scope of the sea clutter region. At the same distance, it becomes a sea clutter region in high sea state, and may become a noise region in low sea state. Therefore, it is unreasonable to treat it as a sea clutter region uniformly. Therefore, a more ideal design is to first identify whether it is a sea clutter region or a noise region before partitioning according to the grazing angle, and then partition the region that has been determined as a sea clutter region and design the target detector according to the partition.

[0006] The literature [Wang Jing, Research on Radar Target Detection Technology under Complex Clutter Background, Master's Thesis of Harbin Institute of Technology, June 2017] proposed a method for distinguishing strong clutter regions, weak clutter regions, and noise regions using the Kullback-Leibler divergence: First, calculate the difference between the PDF (Probability Density Function) of the signal power in a small range and the PDF of the global signal power in the range-azimuth two-dimensional distribution map of the echo signal to obtain a new image. Each pixel point in this image represents a Kullback-Leibler divergence value, and then this image is segmented by the Otsu method. The disadvantages of this method are as follows: First, the algorithm is complex and the computational amount is huge, which is inconvenient for engineering implementation. Second, it is not applicable to radar systems using STC (Sensitivity Time Control) because this method uses the power of the echo signal as the input to obtain the estimated value of the PDF function. However, in a system using STC, the gain of the receiver changes with the distance, and accordingly, the difference between the local and global PDF functions of the echo power cannot accurately reflect the clutter distribution. In an LPRF (Low Pulse Repetition Frequency) radar system, using STC can reduce the design pressure of the receiver dynamic range and is very common. Therefore, using the echo power as a judgment condition should be avoided as much as possible.

[0007] In view of the deficiencies of the methods described in the foregoing documents, the present invention discloses a method for separating clutter and noise based on the statistical characteristics of signals, and performing refined target detection according to the clutter and noise partition results. Summary of the Invention

[0008] The object of the present invention is to provide a method for performing refined target detection in a clutter and noise background. In the case where a clutter background and a noise background coexist, the first problem to be solved is the segmentation of the clutter area and the noise area. The present invention utilizes the differences in statistical characteristics between clutter signals and noise signals to segment the clutter area and the noise area. Next, target detectors are designed respectively in the clutter area and the noise area to achieve refined target detection and optimize the detection performance of the radar system.

[0009] A refined radar detection method based on clutter characteristic perception, comprising the following steps:

[0010] Step 1: Starting from the range-frequency two-dimensional matrix of radar signals, search for the power peak positions on each range gate;

[0011] The working process of the radar consists of a system of transmission and reception processes, as shown in the appendix: Figure 1 The radar only collects data during the reception process. The th sampling value during the th reception process is denoted as . Denote all the data collected by the radar during the th reception process by a matrix . Each element in the matrix is a certain sampling value. For example, the th sampling value during the th reception process is located at the intersection of the th row and the th column of this matrix. This matrix has rows and columns, as follows: The th row represents the set of all sampling values during the th reception process, and the th column represents the set of the th sampling value during all reception processes. And each reception process samples times. Usually, the column direction of the matrix is called the pulse dimension, and the row direction of the matrix is called the range dimension.

[0012]

[0013] For convenience, hereinafter, is used to represent the th element in the matrix The row element, using to represent the matrix in the th column element. The is weighted (the window function used for weighting can be selected as Hamming window, Hanning window, etc.) and subjected to DFT transformation (Discrete Fourier Transform), when ranges from 1 to After traversing, all the results of the DFT transformation are represented by the matrix where the th column element is the result of the DFT transformation of

[0014]

[0015] The above matrix after DFT transformation characterizes the distribution of the signal with respect to distance and frequency. The column direction of the matrix is usually called the frequency dimension, and each specific position in the frequency dimension corresponds to a frequency bin. The row direction of the matrix is usually called the distance dimension, and each specific value in the distance dimension corresponds to a distance bin.

[0016] Extract the th column element from the above matrix , calculate its power value, and search for the frequency bin where the peak of the power is located. When ranges from 1 to After traversing, the positions of the peak power on all distance bins are obtained.

[0017] Step 2. Statistically analyze the characteristics of the peak power. The specific operations are as follows:

[0018] Step 2.1. Set the first statistical window . For the th distance bin, statistically analyze the occurrence frequency of the positions of the peak power on the centered and a total of distance bins before and after. This is equivalent to constructing a histogram. The abscissa of the histogram is the frequency bin, and the ordinate represents the number of times the peak power appears on each frequency bin within the statistical window ( It is preferably an odd number, but not limited to an odd number).

[0019] Step 2.2. Search for the position of the maximum value in the histogram.

[0020] Step 2.3. Set the second statistical window , and calculate the total number before and after centered on the position of the maximum value in the histogram. The total number of times on a frequency gate; the frequency gate corresponding to the peak position in the histogram is the position where the peak power after correction is located on this range gate.

[0021] Step 2.4: Divide the total number of times in the second statistical window by the length of the first statistical window to obtain the percentage representing the degree of signal aggregation in the frequency dimension near the th range gate;

[0022] For the operations of Steps 2.1 to 2.4 are performed from to to obtain the percentage representing the degree of signal aggregation in the frequency dimension near each range gate (Step 2.4), and the position where the peak power after correction is located on each range gate (Step 2.3). For to copy the result, and for to copy the result.

[0023] Step 3: Divide the clutter area and noise area in the range dimension

[0024] Set the first decision threshold . When the percentage representing the degree of signal aggregation in the frequency dimension near each range gate is greater than this threshold, it is judged as the clutter area in the range dimension; otherwise, it is judged as the noise area in the range dimension.

[0025] Step 4: Calculate the upper and lower boundaries of the clutter

[0026] For each range gate that has been judged as the clutter area in the range dimension, perform the operations in Step 4 to obtain the lower and upper boundaries of the clutter.

[0027] For the range segment judged as the clutter area, start searching on both sides from the position where the peak power after correction is located. When the power shows an inflection point, it is judged as the lower and upper boundaries of the clutter. The specific operation is as follows: Set the third statistical window , set the second decision threshold , starting from the matrix , for the data that has been judged as the clutter area in the range dimension , search downward for the inflection point of the clutter power from the peak position after correction (Step 2.3) in this column of data. The specific criterion is: The power at a certain frequency gate is not greater than the average power value of the frequency gates outside this frequency gate by If it is several times, it is considered that the clutter power has an inflection point, and this frequency gate is judged as the lower boundary of the clutter. Search upward from the corrected peak position (step 2.3) in this column of data. The specific criterion is: the power at a certain frequency gate is not greater than the average power value on the frequency gates other than this frequency gate times of the average power value on the frequency gates, it is considered that the clutter power has an inflection point, and this frequency gate is judged as the upper boundary of the clutter.

[0028] Perform step 4 operation on each range gate that has been judged as the clutter area in the range dimension to obtain the lower boundary and upper boundary of the clutter.

[0029] Step 5: Smooth the lower boundary and upper boundary of the clutter

[0030] Set the fourth statistical window . Extract the lower boundaries of the clutter on consecutive range gates, remove the maximum and minimum values, calculate the mean of the remaining elements, and take the obtained result as the corrected lower boundary. Perform the same operation on the upper boundary to obtain the corrected upper boundary.

[0031] Step 6: Perform clutter-noise segmentation on the range-frequency two-dimensional plane according to the clutter boundaries, and design different detectors according to the segmentation results.

[0032] For the area where the range dimension is the noise segment, design a two-dimensional constant false alarm rate detector; for the area where the range dimension is the clutter segment but outside the corrected lower and upper boundaries of the clutter, design a one-dimensional constant false alarm rate detector; for the area where the range dimension is the clutter segment and is included between the corrected lower and upper boundaries of the clutter, design an adaptive matched filter detector.

[0033] Specifically: in the noise area of the range dimension, all frequency gates are noise, marked as area I here; the part where the range dimension belongs to the clutter area but the frequency dimension belongs to the noise area, marked as area II here, and this part is the part outside the lower and upper boundaries; the part where the range dimension belongs to the clutter area and the frequency dimension also belongs to the clutter area, marked as area III here, and this part is the part between the lower and upper boundaries.

[0034] For the target detection in area I, design a two-dimensional constant false alarm rate (CFAR) detector, as shown in Appendix Figure 9 (a), where the detection unit is the circular symbol ●, the protection unit is the triangular symbol △, and the sample unit is the square symbol ■. The basic operations of constant false alarm rate detection are as follows:

[0035] 1) If holds, it is judged that there is a target on the detection unit;

[0036] 2) If the above equation does not hold, it is determined that there is no target on the detection unit.

[0037] Where is the signal strength on the detection unit, is the signal strength on the sample unit, is the number of samples, is the relative threshold.

[0038] For target detection in area II, a one-dimensional Constant False Alarm Rate (CFAR) detector is designed. As shown in Appendix Figure 9 (b), the samples of the one-dimensional CFAR detector only come from the range dimension, and the remaining operations are the same as those of the two-dimensional CFAR detector.

[0039] For target detection in area III, an Adaptive Matched Filter (AMF) detector is designed.

[0040] AMF is equivalent to a filter bank that performs matched filtering in the column direction (i.e., the pulse dimension) of the matrix . According to the determined upper and lower boundaries of the clutter, searching for the target Doppler frequency within this frequency range can greatly reduce the computational complexity. As shown in Appendix Figure 9 (c), K sampling values are extracted from the pulse dimension of the matrix as the detection input quantity , and sequence data is extracted on both sides to estimate the covariance matrix:

[0041]

[0042] Let p represent the ideal target signal: , is the Doppler frequency value (after clutter and noise segmentation, the frequency value is accurately known for each frequency bin between the lower and upper boundaries of the clutter), is the pulse period. The operations of AMF detection are as follows:

[0043] 1) If holds, it is determined that there is a target on the detection unit;

[0044] 2) If the above equation does not hold, it is determined that there is no target on the detection unit.

[0045] Where is the relative threshold.

[0046] Compared with the prior art, the advantages of the present technology are as follows: 1) The present technology uses histogram to analyze the aggregation degree of the peak values of signal power to judge whether it belongs to clutter background or noise background in the range dimension. The method of removing the maximum and minimum values and then taking the average is used to statistically calculate the lower and upper boundaries of clutter, which is applicable to the situation where there are both noise segments and clutter segments in the range dimension. Moreover, since the specific values of power are not used as the judgment basis, it is also applicable to the systems using STC control. 2) By using the histogram representing the peak power distribution position near a range gate to correct the peak position of clutter power on this range gate, the influence of strong scatterers such as large-sized sea surface ship targets and ground corner reflectors on the peak position of clutter power is effectively eliminated. Therefore, the peak position of clutter obtained according to this method is accurate and reliable. 3) The present technology starts from the range-frequency two-dimensional matrix for analysis and design, with a simple processing flow and small calculation amount. Moreover, the two-dimensional plane is divided into three regions according to the range and frequency dimensions, and detectors are designed according to the characteristics of the detection background in different regions, so as to further optimize the radar detection performance.

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

[0048] The present invention proposes a refined target detection method under clutter background. According to the statistical characteristics of signals on the range-frequency two-dimensional plane, the range dimension is divided into noise segments and clutter segments. For the noise segments in the range dimension, a two-dimensional constant false alarm rate detector is designed for target detection; for the clutter segments in the range dimension, the lower and upper boundaries of clutter are searched. For the parts outside the lower and upper boundaries, a one-dimensional constant false alarm rate detector is designed; for the parts within the lower and upper boundaries, an adaptive matched filtering detector is designed. The clutter-noise segmentation method provided by the present invention completely starts from the statistical characteristics of signals in radar data, with strong adaptability, accuracy and reliability; the method of designing detectors according to the clutter-noise segmentation results by region can effectively improve the radar detection performance; moreover, the calculation amount of the present technology is small, and the two-dimensional range-frequency matrix is an intermediate quantity in the conventional radar signal processing process. The clutter-noise segmentation algorithm based on this is easy to be implemented in engineering. Therefore, the clutter-noise segmentation and region-based target detection method provided by the present invention has good prospects for engineering promotion. Description of the Drawings

[0049] Figure 1 It is a schematic diagram of the radar transmitting and receiving work process.

[0050] Figure 2 It is the processing flow of refined target detection.

[0051] Figure 3 It is an example of the range-frequency distribution of radar signal power.

[0052] Figure 4Example of the power peak distribution diagram for each distance from the door.

[0053] Figure 5 is Figure 4 Example of a histogram of the peak power distribution positions within a small distance centered on a certain distance from the door in

[0054] Figure 6 Example of the percentage of peak aggregation in the histogram.

[0055] Figure 7 Examples of the lower and upper boundaries of clutter.

[0056] Figure 8 Example of the clutter noise segmentation result.

[0057] Figure 9 Schematic diagrams of three detector structures. Detailed implementation manners

[0058] The technical solutions provided by the present invention will be described in detail below in conjunction with specific embodiments. It should be understood that the following specific implementation manners are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0059] During the operation of a pulsed radar, the transmitting and receiving states alternate, as shown in the appendix Figure 1 : In the receiving state, the radar collects and stores data. Assume that the radar stores sampling values during each receiving process. The data stored during consecutive receiving processes constitutes a matrix with rows and columns. The element at the intersection of the row and the column is the th sampling value during the th receiving process, as shown below: Usually, the column direction of the matrix is called the pulse dimension, and the row direction of the matrix is called the distance dimension.

[0060]

[0061] The processing flow for performing refined target detection using the above matrix as the input is as shown in the appendix Figure 2 and is specifically divided into the following steps:

[0062] Step 1: Calculate the power spectrum

[0063] Perform the following operations on each column of data in the above matrix: Take the th column of data as an example, First, multiply by a window function to achieve weighting. The result of weighting is that the values of the elements in the data become smaller at both ends and larger in the middle, and then perform a DFT transformation. In the column data, a total of times of DFT transformation are performed, and the results form a new matrix , as follows: The elements in the th column are the results of performing DFT transformation. The column direction of the matrix is usually called the frequency dimension, and the row direction of the matrix is usually called the range dimension.

[0064]

[0065] Each element of the above matrix is a complex number. After multiplying each element by its conjugate, the two-dimensional range-frequency distribution of the signal power is obtained, as shown in the appendix Figure 3 . In the figure, the abscissa represents the range, the ordinate represents the frequency, Rmin is the minimum range, Rmax is the maximum range, and PRF is the Pulse Repeation Frequency.

[0066] Step 2: Search for the peak positions of the power spectra on each range gate and count the degree of concentration of the peak positions near each range gate.

[0067] Specifically, for each column of elements in the appendix Figure 3 , search for the position of the peak power point, and the obtained results are as shown in the appendix Figure 4 . For each range gate in the appendix Figure 4 , perform the following operations: Count the positions of the peak power points on the range gates centered on this range gate and covering a total of range gates before and after ( it is better to take an odd number, but it is not limited to an odd number). Taking as an example, count the positions of the peak power points near a certain range gate in the appendix Figure 4 , and the obtained results are represented by a histogram as shown in the appendix Figure 5 . After performing the same operations on all range gates, each range gate in the appendix Figure 4 will obtain a histogram as shown in the appendix Figure 5 . During this process, for the 1st to the th range gates, copy the results of the th range gate, and for the th to the th range gates, copy the results of the th range gate.

[0068] Step 3: Divide the clutter area and the noise area in the range dimension. Meanwhile, correct the peak position of the power spectrum in the clutter area.

[0069] For the histogram at each range bin, search for the maximum value, and then count the total percentage within a total of frequency bins centered at this point, forward and backward. The resulting data is shown in Appendix Figure 6 as shown, and Appendix Figure 6 uses as an example. Then set the first decision threshold . When the percentage in Appendix Figure 6 exceeds , it is judged as the clutter area in the range dimension; otherwise, it is judged as the noise area. Appendix Figure 6 uses as an example. For the part judged as the clutter area, use the position of the maximum value in the histogram at this range bin as the corrected value of the power spectrum peak position.

[0070] Step 4: Search for the lower and upper boundaries of the clutter and smooth the lower and upper boundaries of the clutter

[0071] Set the second decision threshold , the third statistical window . For the data at each range bin in Appendix Figure 3 , starting from the corrected peak position mentioned above, perform the following operations one frequency bin downward for each frequency bin: Calculate the average power of the signals in frequency bins outside this frequency bin. If the power at this frequency bin is greater than times the average power, slide to the next frequency bin and continue the same operation; otherwise, it is considered that the inflection point of the signal power has occurred and is judged as the lower boundary of the clutter. Similarly, starting from the corrected peak position mentioned above, search one frequency bin upward for each frequency bin to obtain the upper boundary of the clutter.

[0072] Smooth the lower and upper boundaries obtained through the above search. The specific operation is as follows: Set the fourth statistical window . For any range bin, extract the upper boundaries at a total of range bins centered at this range bin, forward and backward. Exclude the maximum and minimum values, and calculate the mean of the remaining elements, which is used as the corrected result of the upper boundary. For the lower boundary, perform smoothing through the same operation. The smoothed results of the lower and upper boundaries of the clutter in Appendix Figure 3 are shown in Appendix Figure 7 as shown, and Appendix Figure 7 uses as an example.

[0073] Step 5: Perform clutter-noise segmentation on the range-frequency two-dimensional plane according to the clutter boundaries.

[0074] According to the appendix Figure 6 The percentages in it are segmented from the range dimension. For the area that does not exceed the threshold is divided into the noise segment; otherwise, it is divided into the clutter area. For the part divided into the clutter area from the range dimension, next, it is divided according to the lower and upper boundaries of the clutter. If it is outside the lower and upper boundaries, it is divided into the noise area; if it is in the middle of the lower and upper boundaries, it is divided into the clutter area.

[0075] Step 6: Perform partition detection according to the partition result

[0076] The final result of the segmentation in the two-dimensional plane of range-frequency is as shown in the appendix Figure 8 : In the noise area of the range dimension, it is all noise on all frequency gates, marked as Area I here; for the part where the range dimension belongs to the clutter area but the frequency dimension belongs to the noise area, it is marked as Area II here, and this part is the part outside the lower and upper boundaries; for the part where the range dimension belongs to the clutter area and the frequency dimension also belongs to the clutter area, it is marked as Area III here, and this part is the part in the middle of the lower and upper boundaries.

[0077] For the target detection in Area I, design a two-dimensional Constant False Alarm Rate (CFAR) detector, as shown in the appendix Figure 9 (a), where the detection unit is the circular symbol ●, the protection unit is the triangular symbol △, and the sample unit is the square symbol ■. The basic operation of the constant false alarm rate detection is as follows:

[0078] 1) If holds, then it is determined that there is a target on the detection unit;

[0079] 2) If the above formula does not hold, then it is determined that there is no target on the detection unit.

[0080] Where is the signal strength on the detection unit, is the signal strength on the sample unit, is the number of samples, is the relative threshold.

[0081] For the target detection in Area II, design a one-dimensional Constant False Alarm Rate (CFAR) detector, as shown in the appendix Figure 9 (b). The samples of the one-dimensional constant false alarm rate detector only come from the range dimension, and the rest of the operations are the same as those of the two-dimensional constant false alarm rate detector.

[0082] For the target detection in Area III, design an Adaptive Matched Filter (AMF) detector.

[0083] The AMF is equivalent to a filter bank that performs matched filtering in the column direction of the matrix (i.e., the pulse dimension). According to the determined clutter upper and lower boundaries, searching for the target Doppler frequency within this frequency range can greatly reduce the computational amount. As shown in Appendix Figure 9 (c), extract K sampling values in the pulse dimension of the matrix as the detection input quantity , and extract sequence data on both sides to estimate the covariance matrix:

[0084]

[0085] Denote the ideal target signal by p: , is the Doppler frequency value (after clutter and noise segmentation, the frequency value is precisely known for each frequency bin between the lower and upper boundaries of the clutter), is the pulse period. The operation of AMF detection is as follows:

[0086] 1) If holds, then it is determined that there is a target on the detection unit;

[0087] 2) If the above formula does not hold, then it is determined that there is no target on the detection unit.

[0088] where is the relative threshold.

[0089] As described above, the above is only the best specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

[0090] The content not described in detail in the specification of the present invention belongs to the well-known technology of those skilled in the art.

Claims

1. A refined radar detection method based on clutter characteristic perception, characterized in that, the steps of this method are as follows: Step 1: Starting from the range-frequency two-dimensional matrix of radar signals, search for the power peak positions on each range gate; Step 2: Statistically analyze the characteristics of the peak power. The specific operations are as follows: Step 2.1: Set the first statistical window W 1 , for the m-th range gate, count the occurrence frequency of the positions of the peak powers on the range gates that are centered around m and have a total of W 1 range gates before and after, which is equivalent to constructing a histogram. The abscissa of the histogram is the frequency gate, and the ordinate represents the number of times the peak power within the statistical window W 1 appears on each frequency gate. It is preferably odd for W 1 , but it is not limited to odd numbers; Step 2.2: Search for the position where the maximum value in the histogram is located; Step 2.3, set the second statistical window W 2 , calculate the total number of times in the histogram within a total of W 2 frequency gates centered on the position of the maximum value; the frequency gate corresponding to the peak position in the histogram is the position of the corrected peak power on this range gate; Step 2.

4. Divide the total number of times in the second statistical window W 2 by the length of the first statistical window W 1 to obtain the percentage representing the degree of signal aggregation in the frequency dimension near the m-th range gate; For m ranging from m = (W 1 +1) / 2 to m = M - (W 1 -1) / 2, perform operations in steps 2.1 to 2.4 to obtain the percentage representing the degree of signal aggregation in the frequency dimension near each range gate, and the position of the peak power on each range gate after correction; for m = 1 to m = (W 1 -1) / 2, copy the result of m = (W 1 +1) / 2, and for m = M - (W 1 -3) / 2 to m = M, copy the result of m = M - (W 1 -1) / 2; Step 3: Divide the clutter area and noise area in the range dimension Step 4: Calculate the upper and lower boundaries of the clutter Perform the operation of Step 4 on each range gate that has been determined to be in the clutter area of the range dimension to obtain the lower and upper boundaries of the clutter; Step 5: Smooth the lower and upper boundaries of the clutter Step 6: Perform clutter-noise segmentation on the range-frequency two-dimensional plane according to the clutter boundaries, and design different detectors according to the segmentation results; In the above Step 1, all the data collected by the radar during N reception processes is represented by a matrix S. Each element in the matrix is a certain sampling value. The m-th sampling value s(n,m) during the n-th reception process is located at the intersection of the n-th row and the m-th column of this matrix. This matrix contains N rows and M columns. The n-th row represents the set of all sampling values during the n-th reception process, and the m-th column represents the set of the m-th sampling value during all reception processes. And each reception process samples M times; the column direction of the matrix S is called the pulse dimension, and the row direction of the matrix S is called the range dimension; Use S(n,:) to represent the elements in the n-th row of the matrix S, and use S(:,m) to represent the elements in the m-th column of the matrix S; Weight S(:,m) and perform DFT transformation. After m traverses from 1 to M, all the results of the DFT transformation are represented by a matrix X. Among them, the m-th column element X(:,m) is the result of the DFT transformation of S(:,m), as follows: The above matrix X after DFT transformation characterizes the distribution of the signal with respect to range and frequency. The column direction of the matrix is usually called the frequency dimension. Each specific position in the frequency dimension corresponds to a frequency gate. The row direction of the matrix is usually called the range dimension. Each specific value in the range dimension corresponds to a range gate; Extract the m-th column element X(:,m) from the above matrix X, calculate its power value, and search for the frequency gate where the peak of the power is located. After m traverses from 1 to M, obtain the positions of the peak powers on all range gates; In step 4, for the distance segment determined as the clutter area, search on both sides starting from the position of the peak power after correction. When the power shows an inflection point, it is determined as the lower and upper boundaries of the clutter. The specific operation is as follows: Set the third statistical window W 3 , set the second decision threshold T 2 . Starting from matrix X, for the data X(:,m) that has been determined as the clutter area in the distance dimension, search downward from the corrected peak position in this column of data for the inflection point of the clutter power. The specific criterion is: The power at a certain frequency gate is not greater than T 3 times the average power value of the W 2 frequency gates outside this frequency gate. Then it is considered that the clutter power shows an inflection point, and this frequency gate is determined as the lower boundary of the clutter; Search upward from the corrected peak position in this column of data. The specific criterion is: The power at a certain frequency gate is not greater than T 3 times the average power value of the W 2 frequency gates outside this frequency gate. Then it is considered that the clutter power shows an inflection point, and this frequency gate is determined as the upper boundary of the clutter; In the above Step 6, in the noise area of the range dimension, there is noise on all frequency gates, which is marked as area I here; the part where the range dimension belongs to the clutter area but the frequency dimension belongs to the noise area is marked as area II here. This part is the part outside the lower and upper boundaries; the part where the range dimension belongs to the clutter area and the frequency dimension also belongs to the clutter area is marked as area III here. This part is the part between the lower and upper boundaries; For target detection in area I, design a two-dimensional constant false alarm rate detector; for target detection in area II, design a one-dimensional constant false alarm rate detector; for target detection in area III, design an adaptive matched filter detector.

2. According to the method described in claim 1, characterized in that, In step 3, set the first decision threshold T 1 , when the percentage representing the degree of signal aggregation in the frequency dimension near each range gate is greater than this threshold, it is judged as the clutter area in the range dimension; otherwise, it is judged as the noise area in the range dimension.

3. According to the method described in claim 1, characterized in that, In step 5, set the fourth statistical window W 4 ; extract the clutter lower boundaries on the door for consecutive W 4 ones, remove the maximum and minimum values, calculate the mean of the remaining elements, and use the obtained result as the corrected lower boundary; perform the same operation on the upper boundary to obtain the corrected upper boundary.

4. According to the method described in claim 1, characterized in that, For the target detection in Area I, a two-dimensional constant false alarm rate detector is designed, where the detection unit is a circular symbol ●, the guard unit is a triangular symbol △, and the sample unit is a square symbol ■. The basic operations of the constant false alarm rate detection are as follows: 1) If holds, then there is a target on the decision detection unit; 2) If the above equation does not hold, it is determined that there is no target on the detection unit; where D is the signal intensity on the detection unit, x 1 , x 2 , L, x K is the signal intensity on the sample unit, K is the number of samples, T 3 is the relative threshold.

5. According to the method described in claim 1, characterized in that, For the target detection in Area II, a one-dimensional constant false alarm rate detector is designed. The samples of the one-dimensional constant false alarm rate detector only come from the range dimension, and the remaining operations are the same as those of the two-dimensional constant false alarm rate detector.

6. According to the method described in claim 1, characterized in that, For the target detection in Area III, an adaptive matched filter detector is designed; The adaptive matched filtering detector is equivalent to a filter bank that performs matched filtering in the column direction of matrix S, i.e., the pulse dimension. According to the determined clutter upper and lower boundaries, searching for the target Doppler frequency within this frequency range can greatly reduce the computational amount. Extract K sampling values in the pulse dimension of matrix S as the detection input quantity z, and extract sequence data z on both sides k Estimate the covariance matrix: Let \(p\) denote the ideal target signal: \(p = [1, \exp(j2\pi f d T r ), L, \exp(j2\pi f d (N - 1)T r )] T , where \(f d \) is the Doppler frequency value and \(T r \) is the pulse period. The operation of the adaptive matched filter detector is as follows: 1) If is established, there is a target on the judgment detection unit; 2) If the above equation does not hold, it is determined that there is no target on the detection unit; where T 4 is the relative threshold.

Images