Sea surface weak target detection method based on Doppler spectrum

Abstract

The invention discloses a sea surface weak target detection method based on Doppler spectrum. The method comprises the following steps: acquiring a radar echo signal; wherein the distance unit where the target is located is known, the distance unit adjacent to the distance unit serves as a protection unit, and the other distance units serve as pure sea clutter units; the distance unit where the target is located serves as a unit to be detected, the pure sea clutter unit serves as a reference unit, and the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit are calculated; determining a median function, variation and median normalization Doppler amplitude spectrum of the sea clutter Doppler amplitude spectrum; calculating a non-zero interval length and a vector entropy, performing global accumulation to serve as feature vectors of sea clutters, and constructing a training set; based on the training set and in combination with a preset false alarm rate, constructing a detector by using a fast convex hull algorithm; after a radar echo signal of an unknown target is obtained, for each distance unit, whether a target exists in a signal sequence corresponding to the distance unit is judged through a constructed detector.

Description

Technical Field

[0001] The invention belongs to the field of radar signal processing, and in particular relates to a sea surface weak target detection method based on Doppler spectrum. Background Art

[0002] For high-resolution surface surveillance radar, detecting weak targets at sea is an important but difficult task. The difficulties mainly come from three aspects: first, the sea clutter received by high-resolution surface surveillance radar has strong non-Gaussian characteristics, which is difficult to model into an accurate statistical model; second, these weak targets with small radar scattering cross-section and weak radar echo are easily submerged in the strong clutter background with sea spikes; third, these weak targets interact with the waves and have complex motion states, making it difficult to parameterize the echo with the target.

[0003] Feature-based detection methods are an important means to effectively detect such targets. Such methods use a combination of multiple differential features and learning algorithms to detect targets. Existing feature-based detection methods are mainly divided into two categories, one is dedicated to the extraction of differential features, and the other is dedicated to the design of detection algorithms. The current focus is on the extraction of differential features. Since the existing methods have high requirements for the signal-to-noise ratio, when faced with targets with low or even negative signal-to-noise ratio and radial velocity, the detection effect is difficult to meet the requirements. In addition, the radar system requires real-time processing capabilities. The detection effect is good but the calculation is complex and it is difficult to meet actual needs. Therefore, it is necessary to study a method that can efficiently detect targets with low or even negative signal-to-noise ratio and radial velocity. Summary of the invention

[0004] The purpose of the present invention is to provide a method for detecting weak targets on the sea surface based on Doppler spectrum, so as to overcome the problem that the existing technology is limited by the signal-to-noise ratio and the requirement of real-time processing of the radar system, and the detection effect cannot meet the requirements when facing a moving target with a low signal-to-noise ratio and radial velocity.

[0005] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0006] A method for detecting weak targets on the sea surface based on Doppler spectrum, comprising:

[0007] Acquire a radar echo signal; wherein the distance unit where the target is located in the radar echo signal is known, and the distance units adjacent to the distance unit are used as protection units, and the remaining distance units are used as pure sea clutter units;

[0008] The distance unit where the target is located is taken as the unit to be detected, and the pure sea clutter unit is taken as the reference unit, and the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit are calculated;

[0009] Using the Doppler amplitude spectrum of the reference unit, the median function of the Doppler amplitude spectrum of the sea clutter is calculated;

[0010] Determine the variation based on the Doppler amplitude spectrum of the reference unit and the median function of the Doppler amplitude spectrum of the sea clutter;

[0011] The median normalized Doppler amplitude spectrum is calculated by using the median function of the Doppler amplitude spectrum of the variation and reference unit and the Doppler amplitude spectrum of the sea clutter;

[0012] The first characteristic maximum non-zero interval length of each sea clutter is calculated by using the median normalized Doppler amplitude spectrum and the zero point; the median normalized Doppler amplitude spectrum is further normalized, and the second characteristic vector entropy of each sea clutter is calculated based on the normalized result; the third characteristic global accumulation of each sea clutter is calculated using the median normalized Doppler amplitude spectrum;

[0013] Calculate the feature vector of each sea clutter in all reference cells respectively and construct a training set;

[0014] Based on the training set and in combination with a preset false alarm rate, a detector is constructed using a fast convex hull algorithm;

[0015] After acquiring the radar echo signal of the unknown target, for each range unit, the constructed detector is used to determine whether there is a target in the signal sequence corresponding to the range unit.

[0016] Further, the calculation of the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit is expressed as:

[0017]

[0018] Where Z(f d ), Z p (f d ) represent the Doppler amplitude spectrum of the unit to be detected and the reference unit, respectively, z(n), z p (n) represents the signal sequence corresponding to the unit to be detected and the reference unit in the radar echo signal, n represents the nth sampling point, f d is the Doppler frequency, T r is the pulse repetition period, p represents the pth reference unit, p = 1, 2, …, P, P is the number of reference units.

[0019] Further, the use of the Doppler amplitude spectrum of the reference unit to calculate the median function of the sea clutter Doppler amplitude spectrum includes:

[0020]

[0021] Where Z p (f d) is the Doppler amplitude spectrum of the reference unit, medain{·} means to find the median value;

[0022] The determining of the variation based on the Doppler amplitude spectrum of the reference unit and the median function of the sea clutter Doppler amplitude spectrum includes:

[0023]

[0024] in, is the median function of the sea clutter Doppler amplitude spectrum, For deterioration;

[0025] The median normalized Doppler amplitude spectrum is calculated by using the median function of the Doppler amplitude spectrum of the variation and reference unit and the Doppler amplitude spectrum of the sea clutter, which is expressed as:

[0026]

[0027] Where max{·} means finding the maximum value.

[0028] Furthermore, the maximum non-zero interval length NZL is expressed as follows:

[0029]

[0030] in represents the median normalized Doppler amplitude spectrum Y(f d ) corresponds to the maximum value of the Doppler frequency, δ1 and δ2 represent Y(f d )exist The nearest zero points on both sides of express The maximum non-zero interval to which it belongs, length{·} represents the length of the sequence.

[0031] Furthermore, the expression of the vector entropy VE is as follows:

[0032]

[0033] in, It represents the normalized maximum and minimum values ​​of Y(f d ), max(f d )、min(f d ) are the Doppler frequencies f d The maximum and minimum values ​​of .

[0034] Furthermore, the expression of the global accumulation GA is as follows:

[0035]

[0036] Furthermore, based on the training set and in combination with a preset false alarm rate, a detector is constructed using a fast convex hull algorithm, which is expressed as:

[0037]

[0038] Where Ω represents the distribution of the training set S in the feature space, |Ω| represents the volume of the feature space, min(·) represents the minimum value, #{·} represents the number of elements in the set, r represents the rth sea clutter, P f is the false alarm rate, [·] indicates rounding.

[0039] Furthermore, after the radar echo signal of the unknown target is acquired, for each range unit, determining whether there is a target in the signal sequence corresponding to the range unit by means of a constructed detector includes:

[0040]

[0041] Where Q represents the number of triangular faces on the three-dimensional convex hull surface of the detector, (v q 1 ,v q 2 ,v q 3 ) represents the three vertices of the qth triangular face of the three-dimensional convex hull, det(·) represents the calculation of the determinant, and max{·} represents the maximum value; ξ represents the eigenvector of each echo in the distance unit; when the statistic η≤0, it means that the eigenvector corresponding to the echo falls inside the three-dimensional convex hull formed by the detector, and the echo is determined to be sea clutter; when the statistic η>0, it means that the eigenvector corresponding to the echo falls outside the three-dimensional convex hull formed by the detector, and the echo is determined to be an echo containing a target;

[0042] After all echoes in each distance unit are determined, it is determined whether there is a target in the detected distance unit based on the proportion of echoes containing the target.

[0043] A target detection device comprises a processor, a memory and a computer program stored in the memory; when the processor executes the computer program, the sea surface weak target detection method based on Doppler spectrum is implemented.

[0044] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the sea surface weak target detection method based on Doppler spectrum is implemented.

[0045] Compared with the prior art, the present invention has the following technical features:

[0046] The present invention overcomes the problem that existing methods cannot achieve a balance between high detection probability and low computational cost for weak targets with low signal-to-clutter ratio and radial velocity. The clutter level is effectively suppressed through median normalization, and difference features are extracted. At the same time, the false alarm rate is accurately controlled based on the convex hull learning algorithm, and the design of the detector is completed, ultimately achieving effective detection of weak targets with low signal-to-clutter ratio and radial velocity at sea at a low computational cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the process of the method of the present invention, wherein the white part is the training process, and the gray part is the testing process in a certain embodiment;

[0048] Figure 2 is the median normalized Doppler amplitude spectrum of sea clutter and echo with target; wherein (a) is the Doppler amplitude spectrum of sea clutter, (b) is the Doppler amplitude spectrum of echo with target, (c) is the median function, (d) is the variation function, (e) is the median normalized Doppler amplitude spectrum of sea clutter, and (f) is the median normalized Doppler amplitude spectrum of echo with target;

[0049] Figure 3 The separability of the three features in the feature space based on the normalized Doppler amplitude spectrum, where (a) is the case at 0.512s of observation time, (b) is the case at 1.024s of observation time, (c) is the case at 2.048s of observation time, and (d) is the case at 4.096s of observation time;

[0050] Figure 4 The convex hull decision region formation process, where (a) is the initial convex hull formed by the pure clutter feature vector in the feature space, and (b) is the convex hull (decision region) and false alarm points that meet the false alarm rate requirements after training;

[0051] Figure 5 The detection performance of the detector proposed in the present invention is compared with the detector based on traditional three features, the detector based on time-frequency three features, and the detector based on SVM on the CSIR radar database in this field;

[0052] Figure 6 The figure compares the detection performance of the detector proposed in the present invention with the detector based on traditional three features, the detector based on time-frequency three features, and the detector based on SVM on the unmanned aerial vehicle (UAV) radar database; among them, (a) is the detector based on traditional three features, (b) is the detector based on time-frequency three features, (c) is the detector based on SVM, and (d) is the detector proposed in the present invention. DETAILED DESCRIPTION

[0053] In view of the fact that existing methods are difficult to detect weak targets with negative signal-to-noise ratio but radial velocity, the difference features (maximum non-zero interval length, vector entropy and global accumulation) extracted in the present invention have low computational complexity and are all spectral features, which are suitable for targets with negative signal-to-noise ratio but radial velocity. Secondly, the present invention uses median and variation for normalization, which can effectively suppress the clutter level and has better robustness to outliers compared to the commonly used mean and variance normalization. The present invention aims to improve the detection probability of weak targets on the sea surface under the premise of accurately controlling the false alarm rate, while reducing the computational cost.

[0054] The present invention provides a method for detecting weak targets on the sea surface based on Doppler spectrum, which is characterized by comprising the following steps:

[0055] Step 1, obtaining a radar echo signal; wherein the distance unit where the target is located in the radar echo signal is known, and the distance units adjacent to the distance unit are used as protection units, and the remaining distance units are used as pure sea clutter units.

[0056] The radar echo signal reflected by the sea surface received by the radar receiver is obtained through the prior database to obtain an M×N dimensional data matrix; N is the number of range units of the radar echo signal, and M is the length of the signal sequence in the range unit; the range unit is the basic unit used to represent the fixed distance range between the target and the radar in the radar echo signal; in this scheme, since the radar echo signal comes from the prior database, the range unit where the target is located is known; the range unit adjacent to the range unit where the target is located is used as a protection unit, and except for the protection unit and the range unit where the target is located, the remaining range units can be considered as pure sea clutter units.

[0057] For example, a detection scene has 11 range units. If the target is located in the 9th range unit, the 8th and 11th range units are protection units, and the remaining range units are pure sea clutter units. Since the echo of the range unit where the target is located will affect the surrounding range units, its neighboring range units are used as protection units.

[0058] Step 2: Take the distance unit where the target is located as the unit to be detected and the pure sea clutter unit as the reference unit, and calculate the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit.

[0059]

[0060] Where Z(f d ), Z p (f d ) represent the Doppler amplitude spectrum of the unit to be detected and the reference unit, respectively, z(n), z p(n) represents the signal sequence corresponding to the unit to be detected and the reference unit in the radar echo signal, n represents the nth sampling point, f d is the Doppler frequency, T r is the pulse repetition period (PRT), p represents the pth reference unit, p = 1, 2, …, P, and P is the number of reference units.

[0061] Step 3, using the Doppler amplitude spectrum of the reference unit, calculate the median function of the sea clutter Doppler amplitude spectrum.

[0062] Assuming that the sea surface is locally uniform, the sea clutter characteristics in the signal sequence of the reference unit are basically consistent with the sea clutter characteristics in the signal sequence of the unit to be detected, so the reference unit can be used to estimate the clutter characteristics. A pure sea clutter unit contains multiple sea clutters, and the number of sea clutters depends on the observation time. For example, taking the 1993IPIX database as an example, when the observation time is 0.512s, there are 1021 sea clutters in each pure sea clutter unit.

[0063] Using the Doppler amplitude spectrum Z of the reference unit p (f d ) Calculate the median function of the sea clutter Doppler amplitude spectrum Specifically expressed as:

[0064]

[0065] Here medain{·} means finding the median.

[0066] Step 4: Determine the variation based on the Doppler amplitude spectrum of the reference unit and the median function of the sea clutter Doppler amplitude spectrum.

[0067]

[0068] Step 5, using the median function of the Doppler amplitude spectrum of the variation and reference unit and the Doppler amplitude spectrum of the sea clutter, calculate the median normalized Doppler amplitude spectrum:

[0069]

[0070] Where max{·} means finding the maximum value.

[0071] Step 6, calculate the first feature of each sea clutter by using the median normalized Doppler amplitude spectrum and the zero point: the maximum non-zero interval length NZL, which is expressed as follows:

[0072]

[0073] in represents the median normalized Doppler amplitude spectrum Y(f d) corresponds to the maximum value of the Doppler frequency, δ1 and δ2 represent Y(f d )exist The nearest zero points on both sides of express The maximum non-zero interval to which it belongs, length{·} represents the length of the sequence.

[0074] Step 7, further normalize the median normalized Doppler amplitude spectrum, and calculate the second feature of each sea clutter based on the normalized result: vector entropy VE, which is expressed as follows:

[0075]

[0076] in, It represents the normalized maximum and minimum values ​​of Y(f d ), max(f d )、min(f d ) are the Doppler frequencies f d The maximum and minimum values ​​of .

[0077] Step 8, use the median normalized Doppler amplitude spectrum to calculate the third feature of each sea clutter: global accumulation GA, which is expressed as follows:

[0078]

[0079] Step 9: Use the first to third features as feature vectors of each sea clutter based on the Doppler amplitude spectrum.

[0080] z r =[NZL,VE,GA] (8)

[0081] Among them, z r Represents the eigenvector of the rth sea clutter in the pure sea clutter unit.

[0082] Step 10, calculate the feature vector of each sea clutter in all reference cells respectively, and construct a training set S.

[0083]

[0084] The eigenvector of the rth sea clutter is represented by z in the above formula: r =[z 1,r ,z 2,r ,z 3,r ], R is the total number of sea clutter contained in all reference cells.

[0085] Step 11, based on the training set and in combination with a preset false alarm rate, a detector is constructed using a fast convex hull algorithm, as follows:

[0086]

[0087] Where Ω represents the distribution of the training set S in the feature space, |Ω| represents the volume of the feature space, min(·) represents the minimum value, #{·} represents the number of elements in the set, r represents the rth sea clutter, P f is the false alarm rate, [·] indicates rounding.

[0088] Step 12, in actual application, after obtaining the radar echo signal of the unknown target, for each distance unit, the constructed detector is used to determine whether there is a target in the signal sequence corresponding to the distance unit. Specifically:

[0089]

[0090] Where Q represents the number of triangular faces on the three-dimensional convex hull surface of the detector, (v q 1 ,v q 2 ,v q 3 ) represents the three vertices of the qth triangular face of the three-dimensional convex hull, det(·) represents the calculation of the determinant, and max{·} represents the maximum value; ξ represents the eigenvector of each echo in the distance unit; when the statistic η≤0, it means that the eigenvector corresponding to the echo falls inside the three-dimensional convex hull formed by the detector, and the echo is determined to be sea clutter; when the statistic η>0, it means that the eigenvector corresponding to the echo falls outside the three-dimensional convex hull formed by the detector, and the echo is determined to be an echo containing a target; when all the echoes in each distance unit are determined, it is determined whether there is a target in the detected distance unit according to the proportion of echoes containing targets.

[0091] Example:

[0092] In one embodiment of the present invention, 10 sets of data from the CSIR radar database and data from the UAV radar database are used; the comparison between the detector proposed in the present invention and the detector based on the traditional three features, the detector based on the time-frequency three features and the detector based on the SVM on the CSIR radar database is as follows: Figure 5 As shown in FIG. 1 , the comparison of the detector proposed in the present invention with the detector based on the traditional three-features, the detector based on the time-frequency three-features and the detector based on SVM on the UAV radar database is shown in FIG. Figure 6 shown.

[0093] (1) In the CSIR radar database, the length of each pulse used is 512 and the false alarm rate is 0.001. Figure 5As shown, in descending order of detection performance, the order is the detector proposed in the present invention, the detector based on the three time-frequency features, the detector based on the traditional three features, and the detector based on SVM. The detector proposed in the present invention obtains the highest average detection probability of 0.902, and the results of the other three detectors are 0.897, 0.755 and 0.693 respectively. In addition, when the results of the proposed detector are equivalent, its computational cost is much lower than that of the detector based on the three time-frequency features with suboptimal performance. In the detector based on the three time-frequency features, the complexity of the smoothed pseudo-Wigner-Ville distribution is O(NKlog2K), where N is the number of pulses and K is the length of the time window. In contrast, the complexity of the proposed Doppler feature is O(N(P+1)), where P is the number of reference units.

[0094] (2) Compared with the results of the CSIR radar database, the detector proposed in this invention has the highest detection result on the UAV radar database, reaching 0.9376, followed by the detector based on the time-frequency three features, the detector based on SVM and the detector based on the traditional three features. Figure 6 In the above example, the target trajectory can only be detected by the detector proposed in this invention and the detector based on the three features of time and frequency.

[0095] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for detecting weak targets on the sea surface based on Doppler spectrum, characterized in that: include: Acquire radar echo signal; The distance unit where the target is located in the radar echo signal is known, and the distance units adjacent to the distance unit are used as protection units, and the remaining distance units are used as pure sea clutter units; The distance unit where the target is located is taken as the unit to be detected, and the pure sea clutter unit is taken as the reference unit, and the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit are calculated; Using the Doppler amplitude spectrum of the reference unit, the median function of the Doppler amplitude spectrum of the sea clutter is calculated; Determine the variation based on the Doppler amplitude spectrum of the reference unit and the median function of the Doppler amplitude spectrum of the sea clutter; The median normalized Doppler amplitude spectrum is calculated by using the median function of the Doppler amplitude spectrum of the variation and reference unit and the Doppler amplitude spectrum of the sea clutter; The length of the first characteristic maximum non-zero interval of each sea clutter is calculated by the median normalized Doppler amplitude spectrum and the zero point; The median normalized Doppler amplitude spectrum is further normalized, and the second eigenvector entropy of each sea clutter is calculated based on the normalized result; The third characteristic global accumulation of each sea clutter is calculated using the median normalized Doppler amplitude spectrum; Calculate the feature vector of each sea clutter in all reference cells respectively and construct a training set; Based on the training set and in combination with a preset false alarm rate, a detector is constructed using a fast convex hull algorithm; After acquiring the radar echo signal of the unknown target, for each range unit, the constructed detector is used to determine whether there is a target in the signal sequence corresponding to the range unit.

2. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: The calculation of the Doppler amplitude spectrum of the unit to be detected and the Doppler amplitude spectrum of the reference unit is expressed as: Where Z(f d ), Z p (f d ) represent the Doppler amplitude spectrum of the unit to be detected and the reference unit, respectively, z(n), z p (n) represents the signal sequence corresponding to the unit to be detected and the reference unit in the radar echo signal, n represents the nth sampling point, f d is the Doppler frequency, T r is the pulse repetition period, p represents the pth reference unit, p = 1, 2, …, P, P is the number of reference units.

3. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: The method of calculating the median function of the sea clutter Doppler amplitude spectrum by using the Doppler amplitude spectrum of the reference unit comprises: Where Z p (f d ) is the Doppler amplitude spectrum of the reference unit, medain{·} means to find the median value; The determining of the variation based on the Doppler amplitude spectrum of the reference unit and the median function of the sea clutter Doppler amplitude spectrum includes: in, is the median function of the sea clutter Doppler amplitude spectrum, For deterioration; The median normalized Doppler amplitude spectrum is calculated by using the median function of the Doppler amplitude spectrum of the variation and reference unit and the Doppler amplitude spectrum of the sea clutter, which is expressed as: Where max{·} means finding the maximum value.

4. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: The expression of the maximum non-zero interval length NZL is as follows: in represents the median normalized Doppler amplitude spectrum Y(f d ) corresponds to the maximum value of the Doppler frequency, δ1 and δ2 represent Y(f d )exist The nearest zero points on both sides of express The maximum non-zero interval to which it belongs, length{·} represents the length of the sequence.

5. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: The expression of the vector entropy VE is as follows: in, It represents the normalized maximum and minimum values ​​of Y(f d ), max(f d )、min(f d ) are the Doppler frequencies f d The maximum and minimum values ​​of .

6. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: The expression of the global accumulation GA is as follows:

7. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: Based on the training set and combined with the preset false alarm rate, a detector is constructed using a fast convex hull algorithm, which is expressed as: Where Ω represents the distribution of the training set S in the feature space, |Ω| represents the volume of the feature space, min(·) represents the minimum value, #{·} represents the number of elements in the set, r represents the rth sea clutter, P f is the false alarm rate, [·] indicates rounding.

8. The method for detecting weak sea targets based on Doppler spectrum according to claim 1, characterized in that: After the radar echo signal of the unknown target is acquired, for each distance unit, a constructed detector is used to determine whether there is a target in the signal sequence corresponding to the distance unit, including: Where Q represents the number of triangular faces on the three-dimensional convex hull surface of the detector, (v q 1 ,v q 2 ,v q 3 ) represents the three vertices of the qth triangular face of the three-dimensional convex hull, det(·) represents the calculation of the determinant, and max{·} represents the maximum value; ξ represents the eigenvector of each echo in the distance unit; when the statistic η≤0, it means that the eigenvector corresponding to the echo falls inside the three-dimensional convex hull formed by the detector, and the echo is determined to be sea clutter; when the statistic η>0, it means that the eigenvector corresponding to the echo falls outside the three-dimensional convex hull formed by the detector, and the echo is determined to be an echo containing a target; After all echoes in each distance unit are determined, it is determined whether there is a target in the detected distance unit based on the proportion of echoes containing the target.

9. A target detection device, comprising a processor, a memory, and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the method for detecting weak sea targets based on Doppler spectrum according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, wherein a computer program is stored in the medium; characterized in that: When the computer program is executed by a processor, the method for detecting weak sea targets based on Doppler spectrum according to any one of claims 1 to 8 is implemented.

Images