Sea surface target detection method and system based on multi-domain features, and equipment medium

Abstract

The invention belongs to the technical field of radar target detection, and particularly relates to a sea surface target detection method and system based on multi-domain features, and an equipment medium. The method comprises the following steps: preprocessing sea clutter unit data, constructing a fluctuation stability feature vector # imgabs0 # time-frequency energy ridge chaos feature vector # imgabs1 # maximum singular value feature vector # imgabs2 # of a sea clutter unit, and combining the fluctuation stability feature vector # imgabs0 # time-frequency energy ridge chaos feature vector # imgabs1 # maximum singular value feature vector # imgabs2 # into a sea clutter feature vector pk; the sea clutter feature vector pk is put into a training set in a target building detector for training to obtain a decision area, data of a to-be-detected unit is preprocessed, the sea clutter feature vector is built, a feature vector pd of the to-be-detected unit is generated, and the feature vector pd is put into the decision area to complete target detection; according to the method, the fluctuation stability, the time-frequency energy ridge chaos degree and the maximum singular value feature are comprehensively analyzed, the target and the sea clutter are strengthened and distinguished, the problem of sample imbalance can be effectively solved in combination with the single-classification support vector machine algorithm, and accurate detection of the floating small target under the sea clutter background is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar target detection, and in particular relates to a sea surface target detection method, system, and device medium based on multi-domain features. Background Art

[0002] Detecting weak targets in the presence of sea clutter is a major challenge in radar remote sensing. Floating targets on the sea surface have a small radar cross section (RCS), low speed, and weak radar echoes, making them easily submerged in sea clutter. The low angle of incidence of the low-grazing radar beam on the sea surface significantly increases the clutter generated by the sea surface. This sea clutter not only approaches or even exceeds the intensity of the target echo, but also has a high degree of similarity in time and frequency to the target echo signal. This poses a significant challenge to target recognition and extraction, especially in strong sea clutter environments. Traditional sea surface target detection methods, which rely on single features or single-domain information for discrimination, have difficulty effectively distinguishing targets from clutter and also struggle to address sample imbalance, significantly impacting the accuracy and reliability of target detection. Therefore, research on sea surface target detection methods based on multi-domain features has important theoretical and practical value.

[0003] In the existing technology, small target feature detection on the sea surface based on frequency domain relative sample entropy (Shi Sainan, Jiang Li, Cao Ding, et al. Journal of Nanjing University of Information Science and Technology (Natural Science Edition), 2023, 15(04): 429-438) uses the frequency domain relative sample entropy (FD-RSE) feature extraction method to perform target detection, which can effectively improve the target detection performance. However, relying on a single frequency domain relative sample entropy, it is difficult to cope with complex ocean environments and diverse target characteristics, and the clutter suppression assumption does not hold in practical applications, which may affect the detection effect.

[0004] In the existing technology, the study of sea surface target detection methods based on time-frequency analysis (Zhang Junling. Xidian University, 2022.) studied three typical time-frequency analysis methods based on time-frequency analysis, and introduced an improved adjustable Q-factor wavelet transform algorithm to improve detection capabilities; however, the study only compared the three methods and did not explore other time-frequency analysis techniques, which limited the universality of the method.

[0005] In the existing technology, the sea surface target detection method based on polarization joint features (Chen Shichao, Gao Heting, Luo Feng. Journal of Radar, 2020, 9(04): 664-673.) uses the polarization covariance matrix and the polarization scattering matrix, combined with Cloude and Krogager feature decomposition and principal component analysis, and adopts support vector machine for target detection; although the polarization characteristics provide a new dimension for distinguishing targets from sea clutter, the single polarization domain feature extraction method fails to fully reflect the characteristics of targets and sea clutter, especially in special sea conditions or when the target characteristics change, which may affect the detection performance. Since the characteristic differences between sea clutter and target echoes in these single domains are not significant, the detection performance is limited and the false alarm rate is high. Summary of the Invention

[0006] To overcome the shortcomings of the above-mentioned prior art, the present invention aims to propose a sea surface target detection method, system, and device medium based on multi-domain features. The target detection method strengthens the distinction between targets and sea clutter through a comprehensive analysis of undulation stability, time-frequency energy ridge chaos, and maximum singular value features. This method can effectively distinguish targets from sea clutter signals. Furthermore, combined with a single-class support vector machine algorithm, it can effectively address the problem of sample imbalance and achieve accurate detection of small floating targets against a sea clutter background.

[0007] To achieve the above object, the technical solutions adopted by the present invention are as follows:

[0008] In a first aspect, a method for detecting sea surface targets based on multi-domain features comprises the following steps:

[0009] S1, performing normalization processing and sliding window processing on the original radar echo data of the sea clutter unit in sequence to obtain expanded data;

[0010] S2, using the expanded data in step S1 to construct the heaving stability characteristic vectors of the sea clutter unit Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector The fluctuation stability characteristic vector Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector Combined into sea clutter eigenvector p k ;

[0011] S3, using a single-class support vector machine (OC-SVM) to build a target detector, the target detector includes a training set and a test set, and the sea clutter feature vector p in step S2 is converted to k Put it into the training set for training to obtain a decision region;

[0012] S4, repeating steps S1 to S2, wherein the original radar echo data of the sea clutter unit in step S1 is replaced by the original radar echo data of the unit to be detected, and the generated feature vector p of the unit to be detected is replaced by the original radar echo data of the unit to be detected. d The target is placed in the decision region described in step S3 to complete target detection.

[0013] Furthermore, the heaving stability characteristic vector of the sea clutter unit in step S2 is The build process consists of the following steps:

[0014] S2.1, the expanded data x in step S1.1 i,j [n] performs standard deviation calculation, and the calculation formula is as follows:

[0015] P1=std(x i,j [n])

[0016] Among them, P1 is the characteristic value of fluctuation stability;

[0017] S2.2. Extract each expanded data x in step S2.1 i,j [n] corresponds to P1 and forms the fluctuation stability characteristic vector

[0018] Furthermore, the time-frequency energy ridge chaos characteristic vector of the sea clutter unit in step S2 is The build process consists of the following steps:

[0019] S2.3, the expanded data x in step S1.1 i,j [n] performs short-time Fourier transform, and the calculation formula is as follows:

[0020]

[0021] in, represents the time-frequency matrix, t represents the time axis sampling point, f represents the frequency axis sampling point, x(m) represents the input signal, h represents the window function, m represents the sampling point in the time window, represents the complex exponential basis (i.e., the rotation factors of the Fourier transform);

[0022] S2.4. Modulate the frequency dimension of the STFT (t, f) obtained by the short-time Fourier transform in step S2.3 and perform the maximum indexing. The calculation formula is as follows:

[0023] V = arg max {|STFT(t,f)|}

[0024] Where V represents the maximum value index;

[0025] S2.5. Perform standard deviation calculation on the maximum value index V described in step S2.4. The calculation formula is as follows:

[0026] P2=std(V)

[0027] Among them, P2 represents the characteristic value of the time-frequency energy ridge disorder;

[0028] S2.6. Extract each expanded data x described in step S2.5 i,j [n] corresponds to P2 and forms the time-frequency energy ridge disorder characteristic vector

[0029] Furthermore, the maximum singular value eigenvector of the sea clutter unit in step S2 is The build process consists of the following steps:

[0030] S2.7, the expanded data x in step S1.1 i,j [n] are rearranged to construct the Hankel matrix H. The Hankel matrix H is subjected to singular value decomposition to obtain a diagonal matrix. The calculation formula is as follows:

[0031]

[0032] D=diag(σ1,σ2,σ3,σ4,σ5),σ1≥σ2≥σ3≥σ4≥σ5

[0033] Where S[k] represents the kth sample value of the sequence (k = 1, 2, 3, ..., C, ..., N), H is the Hankel matrix representing the permuted input data sequence, which is formed by rearranging the input signal, where N is the length of the input sequence, C is the number of columns (window width), the matrix has a constant diagonal structure, U and V are the left singular vector matrix and the right singular vector matrix in SVD, respectively, and D represents the singular value diagonal matrix whose diagonal elements are σ1, σ2, σ3, σ4, σ5;

[0034] S2.8. Extract the first singular value in the diagonal matrix described in step S2.7. The calculation formula is:

[0035] P3=σ1

[0036] Among them, P3 represents the maximum singular value eigenvalue;

[0037] S2.9, extract each expanded data x described in step S2.8 i,j [n] corresponds to P3 and forms the largest singular value eigenvector

[0038] Furthermore, the sea clutter feature vector p in step S2 k Satisfies the following formula:

[0039]

[0040] Further, the specific steps for training to obtain the decision region in step S3 are as follows:

[0041] S3.1. Use the sea clutter feature vector p described in step S2 k to train a hyperplane model through a support vector machine algorithm;

[0042] S3.2. Filter the data of the hyperplane model described in step S3.1 with the sea clutter feature vector p according to a preset false alarm rate k so that the distance between the sample points of the sea clutter feature vector p after data filtering in the hyperplane model and the hyperplane model is the largest; k

[0043] S3.3. Output the hyperplane model with the largest distance described in step S3.2 to the test set as the decision region.

[0044] Further, the target detection of the decision region described in step S4 satisfies the following formula:

[0045]

[0046] where N represents the total number of cells in the detection region; α i is the weighting coefficient of the i-th detection cell, used to measure its importance in the overall decision; λ k represents the features or singular values related to the detection data; σ 2 is the estimated variance of the background noise, used for normalizing the eigenvalue; ρ is the detection threshold for controlling the distinction between targets and non-targets; k represents the cut-off index of the main eigenvalue. When k < n, only non-dominant components are considered, used to determine whether it is a sea clutter region; when , the cell to be detected is a target cell, and when , the cell to be detected is sea clutter.

[0047] In the second aspect, a sea surface target detection system based on multi-domain features includes the following modules:

[0048] Preprocessing module: Normalize and perform sliding window processing on the original radar echo data of sea clutter cells to obtain extended data;

[0049] Feature vector construction module: Use the extended data to construct the fluctuation stability feature vector of sea clutter cells the time-frequency energy ridge confusion feature vector the maximum singular value feature vector The fluctuation stability feature vector the time-frequency energy ridge confusion feature vector the maximum singular value feature vector Combined into sea clutter eigenvector p k ;

[0050] Target detector construction module: Target detector is constructed using single-class support vector machine (OC-SVM), which contains training set and test set. k Put it into the training set for training to obtain a decision region;

[0051] Target detection module: replace the original radar echo data of the sea clutter unit in the normalization module with the original radar echo data of the unit to be detected, and use the feature vector in the expanded data construction module in the preprocessing module to finally generate the feature vector p of the unit to be detected d , the feature vector p d Place it in the decision area to complete target detection.

[0052] In a third aspect, an electronic device includes a memory and a processor:

[0053] Memory: used for storing a computer program for implementing the sea surface target detection method based on multi-domain features;

[0054] Processor: used to implement the sea surface target detection method based on multi-domain features when executing the computer program.

[0055] In a fourth aspect, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for detecting sea surface targets based on multi-domain features is implemented.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] This method extracts the features of fluctuation stability, time-frequency energy ridge disorder and maximum singular value through steps S1 to S2. Since these features show significant differences in the feature domain, that is, they have high discrimination, through comprehensive analysis of these features, it is possible to effectively distinguish targets from sea clutter signals, thereby greatly improving the accuracy and reliability of detection.

[0058] (2) This method uses the single-class support vector machine algorithm in steps S3 to S4, combined with the fluctuation stability, time-frequency energy ridge chaos and maximum singular value features to detect targets, which can effectively deal with the problem of sample imbalance. In the case of sample imbalance, these features clearly distinguish the different characteristics of the target and clutter, so that the single-class support vector machine can pay more attention to the key features of the target signal and avoid the dominance of sea clutter, thereby maintaining a high detection performance under the condition of unbalanced samples. Since the single-class support vector machine algorithm only needs to train sea clutter data, the target data can be detected as an abnormal signal. Compared with the binary classification algorithm, it can solve the problem of sample imbalance.

[0059] In summary, this method strengthens the distinction between the target and sea clutter through a comprehensive analysis of the fluctuation stability, time-frequency energy ridge chaos, and maximum singular value characteristics. It can effectively distinguish the target from the sea clutter signals, and then combine with the single-class support vector machine algorithm to effectively deal with the sample imbalance problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is the target detection flowchart of this method.

[0061] Figure 2 1 is a characteristic diagram of the fluctuation stability of sea clutter units at different distances in the embodiment.

[0062] Figure 3 1 is a time-frequency energy ridge disorder characteristic diagram of sea clutter at different distance units in the embodiment.

[0063] Figure 4 is a characteristic diagram of the maximum singular value of sea clutter at different distance units in the embodiment.

[0064] Figure 5 Schematic diagram of the separability of feature vectors in feature space in the embodiment.

[0065] Figure 6 This is a line chart comparing the performance of the target detector in this method with the single feature detector and the single domain feature detector. DETAILED DESCRIPTION

[0066] The following is combined with Figure 1 To the attached Figure 6 The present invention is described in detail with reference to the accompanying drawings and examples.

[0067] like Figure 1 As shown, the present invention extracts three eigenvectors, namely, fluctuation stability, time-frequency energy ridge disorder and maximum singular value, in the time domain, time-frequency domain and singular value domain respectively, and utilizes the three eigenvectors extracted from different domains into the single-classification support vector machine algorithm to construct a target detector with controllable false alarms, and uses the target detector to complete sea surface target detection.

[0068] In a first aspect, a method for detecting sea surface targets based on multi-domain features comprises the following steps:

[0069] S1. Preprocessing: normalizing and sliding window processing are performed on the original radar echo data of the sea clutter unit to obtain expanded data;

[0070] S2, using the expanded data in step S1 to construct the heaving stability characteristic vectors of the sea clutter unit Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector The fluctuation stability characteristic vector Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector Combined into sea clutter eigenvector p k That is, the three-dimensional feature vector;

[0071] S3, constructing a target detector: using a single-class support vector machine (OC-SVM) to construct a target detector, the target detector includes a training set and a test set, and the sea clutter feature vector p in step S2 is converted to k Put it into the training set to train and obtain the decision area; wherein the training set completes the sea clutter feature vector p k Train and obtain the decision region, and output the decision region with a given controllable false alarm rate to the test set; the test set determines the feature vector p of the unit to be detected d Whether it is within the decision-making area;

[0072] S4, target detection: repeat steps S1 to S2, wherein the original radar echo data of the sea clutter unit in step S1 is replaced by the original radar echo data corresponding to the unit to be detected containing sea clutter and the target, and the generated feature vector p of the unit to be detected is replaced by the original radar echo data corresponding to the unit to be detected. d Put it into the decision region described in step S3 to complete the target detection, that is, distinguish the target and sea clutter through the OC-SVM single classification task; if the feature vector p of the unit to be detected is d If it falls into the decision region, i.e. the hyperplane model, it is sea clutter, and if it falls outside the decision region, it is a target unit.

[0073] Furthermore, since the extracted fluctuation stability feature is a scalar value and needs to be expanded into a vector, the normalization process in step S1 includes the following steps:

[0074] S1.1. Normalize the original radar echo data of the sea clutter unit to generate normalized original radar echo data x;

[0075] S1.2. Expand the data sample of the original radar echo data x after normalization in step S1.1 using the sliding window technology. Determine the width N and step size of the sliding window. Let the sliding window with width N slide along the time dimension. After expansion, the data x i,j Each sliding step of [n] satisfies the following formula:

[0076] x i,j [n]=x(Δn·(i-1)+1:Δn·(i-1)+N)

[0077] Where i = 1, 2…N represents the sample number generated in the range unit, j = 1, 2…N represents the range unit number, x represents the normalized raw radar echo data, n represents the range unit, Δn represents the vector length, and N represents the width of the sliding window.

[0078] Furthermore, the fluctuation stability characteristic vector of the sea clutter unit constructed in the time domain in step S2 is The build process consists of the following steps:

[0079] S2.1, the expanded data x in step S1.1 i,j [n] performs standard deviation calculation, and the calculation formula is as follows:

[0080] P1=std(x i,j [n])

[0081] Among them, P1 is the characteristic value of fluctuation stability;

[0082] S2.2. Extract each expanded data x in step S2.1 i,j [n] corresponds to P1 and forms the fluctuation stability characteristic vector

[0083] Furthermore, the time-frequency energy ridge chaos degree eigenvector of the sea clutter unit constructed in the time-frequency domain in step S2 is The build process consists of the following steps:

[0084] S2.3, the expanded data x in step S1.1 i,j [n] performs short-time Fourier transform, and the calculation formula is as follows:

[0085]

[0086] in, represents the time-frequency matrix, t represents the time axis sampling point, f represents the frequency axis sampling point, x(m) represents the input signal, h represents the window function, m represents the sampling point in the time window, Represents the complex exponential basis, i.e., the rotation factor of Fourier transform;

[0087] S2.4. Modulate the frequency dimension of the STFT (t, f) obtained by the short-time Fourier transform in step S2.3 and perform the maximum indexing. The calculation formula is as follows:

[0088] V = arg max {|STFT(t,f)|}

[0089] Where V represents the maximum value index;

[0090] S2.5. Perform standard deviation calculation on the maximum value index V described in step S2.4. The calculation formula is as follows:

[0091] P2=std(V)

[0092] Among them, P2 represents the characteristic value of the time-frequency energy ridge disorder;

[0093] S2.6. Extract each expanded data x described in step S2.5 i,j [n] corresponds to P2 and forms the time-frequency energy ridge disorder characteristic vector

[0094] Furthermore, the maximum singular value eigenvector of the sea clutter unit constructed in the singular value domain in step S2 is The build process consists of the following steps:

[0095] S2.7, the expanded data x in step S1.1 i,j [n] are rearranged to construct the Hankel matrix H. The Hankel matrix H is subjected to singular value decomposition to obtain a diagonal matrix. The calculation formula is as follows:

[0096]

[0097]

[0098] D=diag(σ1,σ2,σ3,σ4,σ5),σ1≥σ2≥σ3≥σ4≥σ5

[0099] Where S[k] represents the kth sample value of the sequence (k = 1, 2, 3, ..., C, ..., N), H is the Hankel matrix representing the permuted input data sequence, which is formed by rearranging the input signal, where N is the length of the input sequence, C is the number of columns, i.e., the window width, and the matrix has a constant diagonal structure. U and V are the left and right singular vector matrices in SVD, respectively, and D represents the singular value diagonal matrix whose diagonal elements are σ1, σ2, σ3, σ4, σ5;

[0100] S2.8. Extract the first singular value in the diagonal matrix described in step S2.7. The calculation formula is:

[0101] P3=σ1

[0102] Among them, P3 represents the largest singular value eigenvalue;

[0103] S2.9. Extract the P3 corresponding to each of the extended data x i,j [n] and form the largest singular value eigenvector

[0104] Furthermore, in the three-dimensional undulation stability - time-frequency energy ridge chaos - largest singular value feature space, the sea clutter feature vector p described in step S2 k satisfies the following formula:

[0105]

[0106] Furthermore, the specific steps for obtaining the decision region described in step S3 include the following steps:

[0107] S3.1. Use the support vector machine (SVM) algorithm in Matlab to train the sea clutter feature vector p described in step S2 k to obtain a hyperplane model;

[0108] S3.2. According to the preset false alarm rate, filter the data of the sea clutter feature vector p described in step S3.1 for the hyperplane model k , remove the corresponding number of feature vectors in the hyperplane model described in step S3.1, and make the distance between the sample points of the sea clutter feature vector p after data filtering in the hyperplane model k the largest from the hyperplane model;

[0109] S3.3. Output the hyperplane model with the largest distance described in step S3.2 to the test set as the decision region for the target and sea clutter.

[0110] Furthermore, the target detection of the decision region described in step S4 satisfies the following formula:

[0111]

[0112] Among them, N represents the total number of cells in the detection region; α i is the weighting coefficient of the i-th detection cell, used to measure its importance in the overall decision; λ k represents the feature or singular value related to the detection data; σ 2 is the estimated variance of the background noise, used to normalize the eigenvalue; ρ is the detection threshold used to control the distinction between the target and non-target; k represents the cut-off index of the main eigenvalue. When k < n, only non-dominant components are considered, used to determine whether it is a sea clutter region; when , the cell to be detected is a target cell, and when When , the unit to be detected is sea clutter.

[0113] In the second aspect, a sea surface target detection system based on multi-domain features includes the following modules:

[0114] Preprocessing module: normalizes and performs sliding window processing on the original radar echo data of the sea clutter unit to obtain expanded data;

[0115] Characteristic vector construction module: Use the expanded data to construct the heaving stability characteristic vectors of the sea clutter unit Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector The fluctuation stability characteristic vector Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector Combined into sea clutter eigenvector p k ;

[0116] Target detector construction module: Target detector is constructed using single-class support vector machine (OC-SVM), which contains training set and test set. k Put it into the training set for training to obtain a decision region;

[0117] Target detection module: replace the original radar echo data of the sea clutter unit in the normalization module with the original radar echo data of the unit to be detected, and use the feature vector in the expanded data construction module in the preprocessing module to finally generate the feature vector p of the unit to be detected d , the feature vector p d Place it in the decision area to complete target detection.

[0118] In a third aspect, an electronic device includes a memory and a processor:

[0119] Memory: used for storing a computer program for implementing the sea surface target detection method based on multi-domain features;

[0120] Processor: used to implement the sea surface target detection method based on multi-domain features when executing the computer program.

[0121] In a fourth aspect, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for detecting sea surface targets based on multi-domain features is implemented; the computer-readable storage medium includes: a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0122] Example

[0123] The IPIX radar echo data used in the embodiment is derived from field measurement data obtained by Professor S. Havkin and his team of McMaster University in Canada in 1993 at Dartmouth. The data parameters are as follows: the radar carrier frequency is 9.3 GHz, the pulse repetition frequency is 1 kHz, the sampling frequency is 10 MHz, the beam width is 0.9 degrees, the pulse width is 200 ns, the radar azimuth is 128.919 degrees, and the polarization mode includes a full polarization four-channel mode (HH, HV, VH, and VV).

[0124] like Figure 2 The figure shows the distribution of radar echo fluctuation stability for target and sea clutter cells in HH polarization mode for data set 40. It can be seen that the fluctuation stability characteristics of the target and sea clutter cells differ significantly. In data file 40, the target is in primary range cell 7 and in secondary range cells 5, 6, and 8. The figure shows that the eigenvalue of the pure sea clutter cell is large, fluctuating between 0.55 and 0.9, while the eigenvalue of the target cell fluctuates around 0.3. The sea clutter cell has a larger fluctuation stability eigenvalue due to the fluctuation of the sea waves. The fluctuation stability characteristics of the target and sea clutter show a clear difference.

[0125] like Figure 3 The figure shows the distribution of the time-frequency ridge chaos of the radar echoes of the target and sea clutter units in the HH polarization mode of data set No. 40. It can be seen that the time-frequency ridge chaos characteristics of the target and sea clutter units are significantly different. In data file No. 40, the target is in the primary range unit 7 and the secondary range units are 5, 6, and 8. It can be seen from the figure that the eigenvalue of the pure sea clutter unit is larger, and the value of the sea clutter unit fluctuates around 1.5 to 3, while the eigenvalue of the target unit fluctuates around 0.5. Since the target floats on the sea surface, its time-frequency ridge line is approximately a straight line, so its chaos value is low. The time-frequency ridge chaos characteristics of the target and sea clutter show obvious differences.

[0126] like Figure 4The figure shows the distribution of the maximum singular values of the radar echoes of the target and sea clutter units in the HH polarization mode of the No. 40 data set. It can be seen that the maximum singular value characteristics of the target and sea clutter units are significantly different. In the No. 40 data file, the target is in the primary range unit 7 and the secondary range units are 5, 6, and 8. It can be seen from the figure that the eigenvalue of the target unit is larger, and the value of the sea clutter unit fluctuates around 25 to 150, while the eigenvalue of the target unit fluctuates around 375. Since the target unit has large scattering energy and floats stably on the sea surface, its maximum singular value can accumulate to around 375. However, due to the small energy of sea clutter, the influence of the short-term sea spike can be ignored as the observation time increases. The maximum singular value characteristics of the target and sea clutter show obvious differences.

[0127] As shown in Table 1, under different data lengths of IPIX measured sea clutter data, the false alarm rate is 1e -3 ,Comparison of the performance of this method with that of a single domain feature detector under HH polarization conditions, Figure 6 To convert Table 1 into a line chart, as shown in Table 1 and Figure 6 As shown in the figure, the results of the IPIX radar measured data set test show that the extracted multi-domain features have good separability in the feature space; compared with the detection algorithms of single features and single domain features, it has higher detection performance and can effectively complete the detection.

[0128] Table 1 Detection probability under different data lengths

[0129]

[0130] It can be seen that the target detection probability of this method is higher than that of the single domain feature detector and the single feature detector, and its performance is higher than that of the single domain feature detector and the single feature detector. Figure 5 As shown in FIG, when the observation time is 0.128 s, the boundary between the sea clutter and the target in the embodiment is clearly separated. Therefore, after being processed by this method, the separability between the sea clutter and the target can be significantly improved, and the detection performance is better. After multiple verifications of the IPIX radar measured sea clutter data set, this method is proved to have high detection performance and strong robustness.

[0131] The working principle of the present invention is as follows:

[0132] This method normalizes the original radar echo data, extracts the time domain fluctuation stability, time-frequency energy ridge chaos, and the maximum singular value in the singular value domain of the sea clutter unit, and constructs a feature matrix. Subsequently, the three eigenvectors are input into a single-class support vector machine for training to obtain a decision region, i.e., a hyperplane model. The three eigenvector matrices of the unit to be detected, including sea clutter and target units, are input into the decision region. The support vector machine algorithm is used to determine whether the target eigenvector is within the decision region, thereby realizing sea surface target detection.

Claims

1. A method for detecting sea surface targets based on multi-domain features, characterized in that: The following steps are involved: S1, performing normalization processing and sliding window processing on the original radar echo data of the sea clutter unit in sequence to obtain expanded data; S2, using the expanded data in step S1 to construct the heaving stability characteristic vectors of the sea clutter unit Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector The fluctuation stability characteristic vector Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector Combined into sea clutter eigenvector p k ; S3, using a single-class support vector machine (OC-SVM) to build a target detector, the target detector includes a training set and a test set, and the sea clutter feature vector p in step S2 is converted to k Put it into the training set for training to obtain a decision region; S4, repeating steps S1 to S2, wherein the original radar echo data of the sea clutter unit in step S1 is replaced by the original radar echo data of the unit to be detected, and the generated feature vector p of the unit to be detected is replaced by the original radar echo data of the unit to be detected. d The target is placed in the decision region described in step S3 to complete target detection.

2. The sea surface target detection method according to claim 1, wherein: The heaving stability characteristic vector of the sea clutter unit in step S2 The build process consists of the following steps: S2.1, the expanded data x in step S1.1 i,j [n] performs standard deviation calculation, and the calculation formula is as follows: P1=std(x i,j [n]) Among them, P1 is the characteristic value of fluctuation stability; S2.

2. Extract each expanded data x in step S2.1 i,j [n] corresponds to P1 and forms the fluctuation stability characteristic vector 3. The sea surface target detection method according to claim 1, wherein: The time-frequency energy ridge chaos characteristic vector of the sea clutter unit in step S2 The build process consists of the following steps: S2.3, the expanded data x in step S1.1 i,j [n] performs short-time Fourier transform, and the calculation formula is as follows: in, represents the time-frequency matrix, t represents the time axis sampling point, f represents the frequency axis sampling point, x(m) represents the input signal, h represents the window function, m represents the sampling point in the time window, represents the complex exponential basis (i.e., the rotation factors of the Fourier transform); S2.

4. Modulate the frequency dimension of the STFT (t, f) obtained by the short-time Fourier transform in step S2.3 and perform the maximum indexing. The calculation formula is as follows: V = argmax{|STFT(t,f)|} Where V represents the maximum value index; S2.

5. Perform standard deviation calculation on the maximum value index V described in step S2.

4. The calculation formula is as follows: P2=std(V) Among them, P2 represents the characteristic value of the time-frequency energy ridge disorder; S2.

6. Extract each expanded data x described in step S2.5 i,j [n] corresponds to P2 and forms the time-frequency energy ridge disorder characteristic vector 4. The sea surface target detection method according to claim 1, wherein: The maximum singular value eigenvector of the sea clutter unit in step S2 The build process consists of the following steps: S2.7, the expanded data x in step S1.1 i,j [n] are rearranged to construct the Hankel matrix H. The Hankel matrix H is subjected to singular value decomposition to obtain a diagonal matrix. The calculation formula is as follows: D=diag(σ1,σ2,σ3,σ4,σ5),σ1≥σ2≥σ3≥σ4≥σ5 Where S[k] represents the kth sample value of the sequence (k = 1, 2, 3, ..., C, ..., N), H is the Hankel matrix representing the permuted input data sequence, which is formed by rearranging the input signal, where N is the length of the input sequence, C is the number of columns (window width), the matrix has a constant diagonal structure, U and V are the left singular vector matrix and the right singular vector matrix in SVD, respectively, and D represents the singular value diagonal matrix whose diagonal elements are σ1, σ2, σ3, σ4, σ5; S2.

8. Extract the first singular value in the diagonal matrix described in step S2.

7. The calculation formula is: P3=σ1 Among them, P3 represents the maximum singular value eigenvalue; S2.9, extract each expanded data x described in step S2.8 i,j [n] corresponds to P3 and forms the largest singular value eigenvector 5. The sea surface target detection method according to claim 1, wherein: The sea clutter feature vector p in step S2 k Satisfies the following formula:

6. The sea surface target detection method according to claim 1, wherein: The training to obtain the decision region in step S3 specifically includes the following steps: S3.1, the sea clutter feature vector p in step S2 k The hyperplane model is obtained by training the support vector machine algorithm; S3.2, according to the preset false alarm rate, the hyperplane model in step S3.1 is subjected to the sea clutter feature vector p k Data filtering makes the sea clutter eigenvector p after data filtering in the hyperplane model k The distance between the sample point and the hyperplane model is the largest; S3.

3. Output the hyperplane model with the largest interval distance described in step S3.2 to the test set as the decision region.

7. The sea surface target detection method according to claim 1, wherein: The target detection in the decision area in step S4 satisfies the following formula: Where, N represents the total number of cells in the detection area; α i is the weighting coefficient of the i-th detection cell, used to measure its importance in the overall decision; λ k represents the feature or singular value related to the detection data; σ 2 is the estimated variance of the background noise, used to normalize the eigenvalue; ρ is the detection threshold for controlling the distinction between the target and non-target; k represents the cut-off index of the main eigenvalue. When k < n, only non-dominant components are considered, used to determine whether it is a sea clutter area; when holds, the cell to be detected is a target cell, and when holds, the cell to be detected is sea clutter.

8. A sea surface target detection system based on multi-domain features, applying the sea surface target detection method according to claim 1, characterized in that: Includes the following modules: Preprocessing module: normalizes and performs sliding window processing on the original radar echo data of the sea clutter unit to obtain expanded data; Characteristic vector construction module: Use the expanded data to construct the heaving stability characteristic vectors of the sea clutter unit Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector The fluctuation stability characteristic vector Time-frequency energy ridge disorder eigenvector Largest singular value eigenvector Combined into sea clutter eigenvector p k ; Target detector construction module: Target detector is constructed using single-class support vector machine (OC-SVM), which contains training set and test set. k Put it into the training set for training to obtain a decision region; Target detection module: replace the original radar echo data of the sea clutter unit in the normalization module with the original radar echo data of the unit to be detected, and use the feature vector in the expanded data construction module in the preprocessing module to finally generate the feature vector p of the unit to be detected d , the feature vector p d Place it in the decision area to complete target detection.

9. An electronic device comprising a memory and a processor, characterized in that: Memory: used for storing a computer program for implementing the sea surface target detection method; Processor: used to implement the sea surface target detection method when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the sea surface target detection method is implemented.

Images