A sea surface small target detection method and system

Abstract

The present application relates to the field of power system automation, disclose a sea surface small target detection method and system, the present application will be based on principal component analysis anomaly detection method and controllable false alarm support vector machine are combined. Formed the overall detector retains the advantages of anomaly detection method, solves the problem of non-exhaustive target signal and sample imbalance. Retains the advantages of controllable false alarm support vector machine control false alarm rate, good detection performance. Experiments show that the algorithm proposed in the present application can be well adapted to the unique scene of sea surface small target detection problem, under the condition of controlling false alarm rate, shows better detection performance.

Description

Technical Field

[0001] This invention relates to the field of signal technology, and in particular to a method and system for detecting small targets on the sea surface. Background Technology

[0002] More than 70% of the Earth's surface is covered by oceans, containing resources and abundant information essential for human survival. Therefore, detecting small targets on the sea surface is crucial for resource development, environmental protection, and maritime transportation. However, detecting small floating targets on the sea surface remains a key challenge in radar signal processing. This is primarily because sea clutter is influenced by various environmental factors, resulting in non-constant characteristics, strong non-Gaussianity, and spatiotemporal correlation. Furthermore, the small size of the target leads to weak echo intensity; moreover, the small size also means that sea conditions have a greater impact on the target, often causing waves to obscure it, making the detection of small targets on the sea surface even more difficult.

[0003] In summary, the difficulties in small target detection on the sea surface mainly lie in two aspects. First, while sea clutter characteristics are ergodic, target characteristics are not. That is, targets are diverse, and it is usually impossible to traverse all targets. Moreover, since sea clutter constitutes a large proportion of the echo signal, an issue that must be considered is the extreme imbalance between the number of sea clutter features and target features. Second, in small target detection on the sea surface, the tolerance for the two types of errors differs significantly. For example, in many cases, the false alarm rate (the probability of misidentifying sea clutter as a target) is typically required to be below 10%. -3 Around 10%. As for the false negative rate (the probability of misidentifying a target as sea clutter), as long as it is below 10%, it is acceptable. -2 The requirements can be met by applying the same penalty metric for both types of errors. However, most existing traditional classifiers use the same penalty metric for both types of errors, which is not well applied to the problem of detecting small targets on the sea surface.

[0004] The best solution to the first problem mentioned above is to introduce a single-class anomaly detection method. This involves treating sea clutter echo features as normal samples and target echo features as anomalous samples, then classifying them using a threshold. Several anomaly detection algorithms have been applied to small target detection on the sea surface, such as K-Nearest Neighbors (KNN) based on Mahalanobis distance, one-class SVM anomaly detection, and Principal Component Analysis (PCA) based anomaly detection. While single-class anomaly detection methods can address the issues of non-ergodic target signals and imbalanced classes, they cannot solve the problem of varying tolerances for false alarms and false negatives. Current anomaly detection methods do not perform well in controlling the false alarm rate (FAR), and their accuracy needs improvement. An effective solution to the second problem is to use a constant or controllable false alarm detector to control the false alarm rate. Several methods have achieved constant false alarm rate (CFAR), such as the CFAR detector and the controllable false alarm support vector machine detector (pf-SVM). These detectors offer advantages such as high accuracy and relatively accurate false alarm control, but they typically require a sufficient number of balanced samples from both classes for training to achieve good detection results.

[0005] Therefore, combining single-class anomaly detection methods with controllable false alarm detection methods to leverage the strengths of both types of detectors and fully utilize their advantages is a pressing technical challenge in the field of small target detection on the sea surface. Summary of the Invention

[0006] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a method and system for detecting small targets on the sea surface, solving the problems of low accuracy when there are different tolerances for false alarms and false negatives, and when the number of samples is insufficient and difficult to balance.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a method for detecting small targets on the sea surface, comprising:

[0009] The radar echo is preprocessed, which includes extracting radar echo signal data and initializing detection algorithm parameters to obtain input data.

[0010] The input data is fed into an improved principal component analysis-based anomaly detector for preliminary detection to obtain a first clutter sample. The preliminary detection includes determining the optimal threshold by using the objective function maximization method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold.

[0011] The first clutter sample is input into a controllable false alarm support vector machine for accurate detection to obtain the second clutter sample.

[0012] The actual false alarm rate is calculated based on the output results of the preliminary detection and the precise detection and compared with the set false alarm rate. The comparison is used to adjust the two penalty parameters. The preliminary detection and precise detection are repeated until the actual false alarm rate is the same as the set false alarm rate, and the optimized penalty parameters are obtained.

[0013] The optimized penalty parameters are input into the overall detector to detect the samples and obtain the classification results.

[0014] As a preferred embodiment of the sea surface small target detection method of the present invention, the extraction of radar echo signal data includes amplitude, fractal, and frequency information;

[0015] The initialization detection algorithm parameters include the penalty intensity C0 for false alarms, the penalty intensity C1 for missed detections, and the upper bound C for the change in the penalty intensity for false alarms. h The lower bound C for the change in the penalty for false alarms. l The compensation parameter k for the penalty severity of false alarms during the anomaly detection phase.

[0016] As a preferred embodiment of the sea surface small target detection method of the present invention, the preliminary detection further includes:

[0017] The temporal entropy (TIE), time hearst exponent (THE), and frequency peak-to-average ratio (FPAR) obtained in the preprocessing are fused into a three-dimensional feature, denoted as Fi, and the training matrix Q is constructed using the three-dimensional feature. Q is an M-row, S-column matrix, and S is the dimension of the feature space of the sample.

[0018] The training matrix Q is standardized to obtain matrix Q. r and Q r The covariance matrix is ​​subjected to singular value decomposition, and the eigenvector matrix U and the eigenvalue diagonal matrix Λ are obtained after decomposition.

[0019] Construct a dimension reduction matrix Y, determine the eigenvectors to be retained in the eigenvector matrix U, and after sorting the eigenvalues ​​from largest to smallest, the eigenvectors corresponding to the first j eigenvalues ​​are used to construct the dimension reduction matrix Y.

[0020] Calculate the reconstruction matrix Z and the reconstruction error e for each sample. m The dimension-reduced matrix Y is reconstructed to its original dimension to obtain the reconstructed matrix Z;

[0021] The reconstruction error is expressed as follows:

[0022]

[0023] Where Z is the reconstruction matrix, m represents a single sample in the training set, the length of the training set is M, and s is a single dimension in the feature space dimension of the sample.

[0024] In a preferred embodiment of the sea surface small target detection method described in this invention, the optimal threshold is determined by maximizing the objective function, and the reconstruction error is iterated from the minimum to the maximum value, with the corresponding reconstruction error used as the threshold Ω each time. q Select the threshold Ω that maximizes the objective function value. q As the optimal threshold Ω, the objective function is expressed as:

[0025]

[0026]

[0027]

[0028] Where fp is the number of false alarms, fn is the number of missed detections, and M is the total number of training set samples. C0 and C1 are the two penalty parameters mentioned in (1.5), and k is a hyperparameter selected by the user.

[0029] As a preferred embodiment of the sea surface small target detection method of the present invention, the step of obtaining the first clutter sample includes: after obtaining the optimal threshold Ω, comparing the reconstruction error of the sample to be tested with the optimal threshold Ω; if the reconstruction error is greater than the optimal threshold Ω, it is judged as a target; otherwise, it is judged as clutter.

[0030] As a preferred embodiment of the sea surface small target detection method of the present invention, wherein: the first clutter sample is input into a controllable false alarm support vector machine for accurate detection to obtain a second clutter sample, including determining a decision hyperplane, denoted as:

[0031]

[0032] stYi[j(ω,Fi)-b]≥1-ξi,i=1,2…,N

[0033] ξi≥0, i=1,2…,N

[0034] Where ξi is a slack variable, and C0 and C1 together constitute the penalty parameter to control the penalty strength for errors. When Yi is +1, the penalty parameter takes C1; when Yi is -1, the penalty parameter takes C0.

[0035] In a preferred embodiment of the sea surface small target detection method described in this invention, the actual false alarm rate is calculated based on the output results of preliminary detection and precise detection, and compared with a set false alarm rate. This comparison is used to adjust two penalty parameters, including...

[0036] If the actual false alarm rate is less than the set false alarm rate, then keep c1 unchanged and decrease C0, which is expressed as:

[0037] C h =C0 and

[0038] If the actual false alarm rate is greater than the set false alarm rate, then keep C1 unchanged and increase C0, which is expressed as:

[0039] C l =C0 and

[0040] Secondly, the present invention provides a system for detecting small targets on the sea surface, comprising,

[0041] The preprocessing module is used to preprocess the radar echo, the preprocessing including extracting radar echo signal data and initializing detection algorithm parameters to obtain input data;

[0042] The preliminary detection module is used to input the input data into an improved principal component analysis-based anomaly detector for preliminary detection to obtain a first clutter sample. The preliminary detection includes determining the optimal threshold by maximizing the objective function method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold.

[0043] The precision detection module is used to input the first clutter sample into a controllable false alarm support vector machine for precision detection in order to obtain the second clutter sample;

[0044] The comparison and judgment module is used to calculate the actual false alarm rate based on the output results of the preliminary detection and the precise detection and compare it with the set false alarm rate. The comparison is used to adjust the two penalty parameters and to cycle through the preliminary detection and precise detection until the actual false alarm rate is the same as the set false alarm rate, so as to obtain the optimized penalty parameters.

[0045] The results classification module is used to input the optimized penalty parameters into the overall detector to detect samples and obtain classification results.

[0046] Thirdly, the present invention provides a computing device, comprising:

[0047] Memory and processor;

[0048] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the small target detection method on the sea surface.

[0049] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for detecting small targets on the sea surface.

[0050] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention combines an anomaly detection method based on principal component analysis with a controllable false alarm support vector machine. The resulting overall detector retains the advantages of the anomaly detection method, solving the problems of non-ergodic target signals and sample imbalance. It also retains the advantages of the controllable false alarm support vector machine, such as accurate control of the false alarm rate and good detection performance. Experiments have shown that the algorithm proposed in this invention can adapt well to the unique scenario of small target detection on the sea surface, exhibiting better detection performance under controlled false alarm rate conditions. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0052] Figure 1 This is a schematic diagram of the overall process of a small target detection method on the sea surface according to an embodiment of the present invention.

[0053] Figure 2 This is a detailed flowchart of the method for detecting small targets on the sea surface according to an embodiment of the present invention, in which the C0 is adjusted to control the false alarm rate. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0058] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0059] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integrated connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0060] Example 1

[0061] Reference Figure 1-2 As an embodiment of the present invention, a method for detecting small targets on the sea surface is provided, comprising:

[0062] S1: Preprocess the radar echo. Preprocessing includes extracting radar echo signal data and initializing detection algorithm parameters to obtain input data.

[0063] Furthermore, radar echo signal data is extracted, including amplitude, fractal, and frequency information;

[0064] Specifically, the extraction process involves segmenting the original radar signal according to the observation time. To obtain more data, overlap between observation times is allowed during segmentation. The segmented signal is denoted as x. In this algorithm, the observation time is set to 1.024s.

[0065] Temporal entropy features are extracted from the segmented signal x. Temporal entropy is calculated using amplitude information from the original echo, thus distinguishing sea clutter from the target. The formula for calculating temporal entropy (TIE) is as follows:

[0066]

[0067] During the extraction process, the amplitude range of x is divided into K groups. Where P(N) k ) represents the probability that the amplitude of the echo sequence falls into the k-th segment, N k This represents the total number of elements that fall into the k-th segment.

[0068] The Time Hearst Exponent (THE) feature is extracted from the segmented signal x. The THE is widely used to describe the fractal characteristics of sea clutter, distinguishing targets from sea clutter by calculating the fractal properties of samples over a certain time scale. The THE extraction process is as follows:

[0069] First, divide the sequence x of length N into L subsequences of equal length, where the length of each subsequence is τ = N / L, and represent the amplitude set of the l-th subsequence as ({x l，1 x l，2 , ..., x l，τ}, l = 1, 2, ..., L), for each subsequence, the mean and standard deviation of its amplitude can be calculated, respectively using and S l This represents the mean and standard deviation of the l-th subsequence.

[0070] Then, using Y l Y represents the cumulative deviation set of subsequence l. l ={Y l，1 Y l，2 , ..., Y l，τ},in Get Y l The range R of the cumulative deviation of subsequence l can then be obtained. l R l =max(Y l，t )-min(Y l，t For each subsequence l (l = 1, 2, ..., L), the corresponding R can be calculated. l Thus, L (R) l / S l Let (R / S) represent these L (R) l / S l The mean of (R / S) represents the time-Hurst exponential characteristic of the echo sequence within the time scale τ.

[0071] The peak-to-average power ratio (PPAR) feature is extracted from the segmented signal x. PPAR utilizes frequency domain transformation, a widely used technique in signal processing, to distinguish clutter from the target using frequency domain information. The formula for calculating PPAR is as follows:

[0072]

[0073] Where N is the length of the segmented signal x, X(z) is the Fourier transform of signal x, and M is the maximum value in X(z).

[0074] The initialization parameters for the detection algorithm include the penalty intensity C0 for false alarms, the penalty intensity C1 for missed detections, and the upper bound C for the change in the penalty intensity for false alarms. h The lower bound C for the change in the penalty for false alarms. l The compensation parameter k for the penalty severity of false alarms during the anomaly detection phase.

[0075] It should be noted that C0 and C1 are important parameters in this algorithm. They can be understood as follows: C0 represents the penalty for false alarms (mistaking sea clutter for a target); C1 represents the penalty for missed detections (mistaking a target for sea clutter). h and C l The parameters used when updating C0, where C h This represents the upper bound of the change in C0, C l This represents the lower bound of the change in C0. In the algorithm, C1 remains constant, and k represents the compensation for the C0 parameter during the anomaly detection phase.

[0076] S2: Input the input data into the improved principal component analysis-based anomaly detector for preliminary detection to obtain the first clutter sample. The preliminary detection includes determining the optimal threshold by using the objective function maximization method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold.

[0077] Furthermore, preliminary tests also included,

[0078] S2.1: The temporal entropy (TIE), time hearst exponent (THE), and frequency peak-to-average ratio (FPAR) features obtained in the preprocessing are fused into a three-dimensional feature, denoted as Fi, and a training matrix Q is constructed using the three-dimensional feature. Q is an M-row, S-column matrix, and S is the dimension of the feature space of the sample.

[0079] Preferably, in this algorithm, S can be 3, and M is the length of the training set.

[0080] S2.2: Standardize the training matrix Q to obtain matrix Q. r and Q rThe covariance matrix is ​​subjected to singular value decomposition, and the eigenvector matrix U and the eigenvalue diagonal matrix Λ are obtained after decomposition.

[0081] S2.3: Construct the dimension reduction matrix Y, determine the eigenvectors to be retained in the eigenvector matrix U, and after sorting the eigenvalues ​​from largest to smallest, use the eigenvectors corresponding to the first j eigenvalues ​​to construct the dimension reduction matrix Y;

[0082] It should be noted that, due to the eigenvalues ​​λ in the eigenvalue diagonal matrix S Each vector corresponds one-to-one with the eigenvectors in U. And λ S The larger the value, the more overall information the corresponding feature retains. Therefore, after sorting the feature values ​​from largest to smallest, the eigenvectors corresponding to the first j feature values ​​are used to construct the dimensionality reduction matrix Y.

[0083] Specifically, the formula for constructing the dimensionality reduction matrix is ​​as follows:

[0084] Y = Q r U s×j

[0085] Among them, Q r The standardized matrix obtained in step S2.2 above, U s×j The new matrix is ​​formed by the eigenvectors corresponding to the j largest eigenvalues ​​of the U matrix obtained in step S2.2, and has a size of s×j.

[0086] S2.4: Calculate the reconstruction matrix Z and the reconstruction error e for a single sample. m The dimension-reduced matrix Y is reconstructed to its original dimension to obtain the reconstructed matrix Z;

[0087] Specifically, the reconstruction matrix Z is represented as:

[0088]

[0089] Where Y is the dimensionality reduction matrix obtained in step S2.3, U mentioned in step S2.3 s×j Transpose of a matrix.

[0090] Calculate the reconstruction error e m Represented as,

[0091]

[0092] Where Z is the reconstruction matrix, m represents a single sample in the training set, the length of the training set is M, and s is a single dimension in the feature space dimension of the sample.

[0093] Furthermore, in S2.5: the optimal threshold is determined by maximizing the objective function. The reconstruction error is iterated from its minimum to its maximum value, and the corresponding reconstruction error is used as the threshold Ω at each step.q Select the threshold Ω that maximizes the objective function value. q As the optimal threshold Ω, the objective function is expressed as:

[0094]

[0095]

[0096]

[0097] Where fp is the number of false alarms, fn is the number of missed detections, and M is the total number of training set samples. C0 and C1 are the two penalty parameters mentioned in (1.5), and k is a hyperparameter selected by the user.

[0098] It should be noted that, to achieve better performance when combined with the other two methods, this invention extends the single penalty parameter in the support vector machine to a double penalty parameter to penalize different errors. Simultaneously, a self-designed objective function improves the threshold selection method for anomaly detection based on principal component analysis, replacing the common algorithm's practice of discarding fixed points with a method that maximizes the objective function.

[0099] On some datasets, the traditional approach of discarding fixed points results in a significant discrepancy between the false alarm rate (FAR) obtained from the test data and the set FAR. This leads to instability in controlling the FAR during the anomaly detection phase, thus reducing the overall FAR control capability of the algorithm. The method described above can effectively adjust the algorithm's emphasis on both types of errors, overcoming the problem of unstable FAR control in traditional principal component analysis-based anomaly detection algorithms, and achieving effective and accurate control of the FAR.

[0100] Meanwhile, the Support Vector Machine (SVM) algorithm is heavily influenced by outlier support vectors during the hyperplane determination process. Inaccurate hyperplane identification can significantly impact detection performance. This invention addresses this by performing anomaly detection before the SVM, filtering out some outlier support vectors in advance and thus improving the overall detection performance. Furthermore, the inclusion of anomaly detection alleviates the difficulties of imbalanced samples and the lack of target traversal in sea target detection, further enhancing the detection performance of this invention.

[0101] Furthermore, acquiring the first clutter sample includes, after obtaining the optimal threshold Ω, comparing the reconstruction error of the sample to be tested with the optimal threshold Ω; if it is greater than the optimal threshold Ω, it is judged as the target; otherwise, it is judged as clutter.

[0102] S3: Input the first clutter sample into a controllable false alarm support vector machine for accurate detection to obtain the second clutter sample;

[0103] Furthermore, the first clutter sample is input into a controllable false alarm support vector machine for accurate detection to obtain the second clutter sample, including determining the decision hyperplane, denoted as:

[0104]

[0105] stYi[k(ω,Fi)-b]≥1-ξi,i=1,2...,N

[0106] ξi≥0, i=1,2…,N

[0107] Where ξi is a slack variable, and C0 and C1 together constitute the penalty parameter to control the penalty strength for errors. When Yi is +1, the penalty parameter takes C1; when Yi is -1, the penalty parameter takes C0.

[0108] It should be noted that the slack variable ξi represents the tolerance for errors. The controllable false alarm support vector machine uses C0 and C1 together to form the penalty parameter to control the penalty for errors. When Yi is +1, the penalty parameter is C1, indicating that in the case of a missed detection, the penalty is controlled by C1. When Yi is -1, the penalty parameter is C0, indicating that in the case of a false alarm, the penalty is controlled by C0.

[0109] After the hyperplane is determined, a controlled false alarm support vector machine performs fine-grained detection. Samples classified as targets, along with those initially classified as targets in S2.5, are all output as targets in the final result of the overall detector. Samples classified as clutter are output as clutter in the final result of the overall detector.

[0110] S4: Calculate the actual false alarm rate based on the output results of the preliminary and precise detections and compare it with the set false alarm rate. This comparison is used to adjust two penalty parameters, including...

[0111] Furthermore, based on the output results of preliminary and precise detections, the actual false alarm rate (PF) is calculated and compared with the set false alarm rate (PF). This comparison is used to adjust two penalty parameters, including...

[0112] S4.1: If the actual false alarm rate is less than the set false alarm rate, keep C1 unchanged and decrease C0, as follows:

[0113] C h =C0 and

[0114] S4.2: If the actual false alarm rate is greater than the set false alarm rate, keep C1 unchanged and increase C0, as follows:

[0115] C l =C0 and

[0116] S4.3: Input the C1 value and the modified C0 value into the improved principal component analysis anomaly detector and the controllable false alarm support vector machine, and iterate through the preliminary detection and precise detection until the actual false alarm rate is the same as the set false alarm rate to obtain the optimized penalty parameters; and record the corresponding C0 value and C1 value when the actual false alarm rate is the same as the set false alarm rate.

[0117] S5: Input the optimized penalty parameters from step S4.3 into the overall detector, which consists of an improved principal component analysis anomaly detector and a controllable false alarm support vector machine, to detect the test samples. This will yield the detection results for the test samples at a given false alarm rate.

[0118] The overall detector of this invention can be trained using a large number of sea clutter samples and a small number of target samples, which can effectively control the false alarm rate and improve the detection performance of the algorithm.

[0119] The above is an illustrative scheme of a method for detecting small targets on the sea surface according to this embodiment. It should be noted that the technical solution of the system for detecting small targets on the sea surface and the technical solution of the method for detecting small targets on the sea surface described above belong to the same concept. For details not described in detail in the technical solution of the system for detecting small targets on the sea surface in this embodiment, please refer to the description of the technical solution of the method for detecting small targets on the sea surface described above.

[0120] The sea surface small target detection system in this embodiment includes:

[0121] The preprocessing module is used to preprocess the radar echo. The preprocessing includes extracting radar echo signal data and initializing detection algorithm parameters to obtain input data.

[0122] The preliminary detection module is used to input the input data into the improved principal component analysis-based anomaly detector for preliminary detection to obtain the first clutter sample. The preliminary detection includes determining the optimal threshold by maximizing the objective function method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold.

[0123] The precision detection module is used to input the first clutter sample into a controllable false alarm support vector machine for precise detection in order to obtain the second clutter sample.

[0124] The comparison and judgment module is used to calculate the actual false alarm rate based on the output results of the preliminary detection and the precise detection and compare it with the set false alarm rate. The comparison is used to adjust the two penalty parameters and to cycle through the preliminary detection and precise detection until the actual false alarm rate is the same as the set false alarm rate, thus obtaining the optimized penalty parameters.

[0125] The results classification module is used to input the optimized penalty parameters into the overall detector to detect samples and obtain classification results.

[0126] This embodiment also provides a computing device suitable for detecting small targets on the sea surface, including:

[0127] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the method for detecting small targets on the sea surface as described in the above embodiments.

[0128] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for detecting small targets on the sea surface as proposed in the above embodiments.

[0129] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for detecting small targets on the sea surface proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0130] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0131] Example 2

[0132] Reference Figure 1-2 As an embodiment of the present invention, a method for detecting small targets on the sea surface is provided. In order to verify its beneficial effects, an exemplary application scenario and a comparative scheme are provided for scientific demonstration.

[0133] See Figure 1 :

[0134] S1: Perform the following preprocessing operations on the raw radar echo to extract information such as amplitude and frequency contained in the echo signal, and initialize parameters such as C0 and C1:

[0135] S1.1: Divide the original radar signal into segments of 1024 points each, shifting each segment 64 points to the right. The resulting signal is denoted as x.

[0136] S1.2: Extract the temporal entropy feature from the segmented signal x. The temporal entropy is calculated using the amplitude information in the original echo, thus distinguishing sea clutter from the target. The formula for calculating temporal entropy (TIE) is as follows:

[0137]

[0138] During the extraction process, the amplitude range of x is divided into K groups. Where P(N) k ) represents the probability that the amplitude of the echo sequence falls into the k-th segment, N k This represents the total number of elements falling into the k-th segment. In our experiment, K = 50.

[0139] S1.2: Extract the Time Hearst Exponent (THE) feature from the segmented signal x. The Time Hearst Exponent is widely used to describe the fractal characteristics of sea clutter. By calculating the fractal properties of a sample over a certain time scale, it distinguishes the target from sea clutter. The Time Hearst Exponent (THE) extraction process is as follows:

[0140] First, divide the sequence x of length N into L subsequences of equal length, where the length of each subsequence is τ = N / L, and represent the amplitude set of the l-th subsequence as ({x l，1 x l，2 , ..., x l，τ}, l = 1, 2, ..., L), for each subsequence, the mean and standard deviation of its amplitude can be calculated, respectively using and S l This represents the mean and standard deviation of the l-th subsequence.

[0141] Then, using Y l Let Y = {Y} represent the cumulative deviation set of subsequence l. l，1 Y l，2 , ..., Y l，τ},in Get Y l The range R of the cumulative deviation of subsequence l can then be obtained. l R l =max(Y l，t )-min(Y l，t For each subsequence l (l = 1, 2, ..., L), the corresponding R can be calculated. l Thus, L (R) l / S l Let (R / S) represent these L (R) l / S lThe mean of (R / S) represents the time-Hurst exponential characteristic of the echo sequence within the time scale τ.

[0142] S1.3: Extract the peak-to-average power ratio (PPAR) feature from the segmented signal x. PPAR utilizes frequency domain transformation, a widely used technique in signal processing, to distinguish clutter from the target. The formula for calculating PPAR is as follows:

[0143]

[0144] Where N is the length of the segmented signal x, X(z) is the Fourier transform of signal x, and M is the maximum value in X(z).

[0145] S1.5: Initialize the parameters used in the algorithm, such as C0, C1, C h C l C0 and C1 are important parameters in this algorithm. C0 represents the penalty for false alarms (mistaking sea clutter for a target); C1 represents the penalty for missed detections (mistaking a target for sea clutter). h and C l The parameters used when updating C0, where C h This represents the upper bound of the change in C0, C l This represents the lower bound of the change in C0. C1 is kept constant in the algorithm. In our experiments, C... h =2, C l =0, C0 is initialized to 1, and C1 is always 1.

[0146] S2: Input the preprocessed data into the improved principal component analysis-based anomaly detector for preliminary detection. The specific steps are as follows:

[0147] S2.1: The obtained TIE, THE, and FPAR features are fused into a three-dimensional feature, denoted as Fi. This feature is then used to construct the training matrix Q. Q is an M-row, S-column matrix, where S is the dimension of the feature space of the samples (in this algorithm, S = 3), and M is the length of the training set.

[0148] S2.2: Standardize the training matrix Q to obtain matrix Q. r and Q r The covariance matrix is ​​subjected to singular value decomposition, and the resulting eigenvector matrix is ​​denoted as U, and the eigenvalue diagonal matrix is ​​denoted as Λ.

[0149] S2.3: Constructing the dimensionality-reduced matrix Y. First, it's necessary to determine the eigenvectors to be retained within the U matrix. Since the eigenvalues ​​λ in the eigenvalue diagonal matrix... S Each vector corresponds one-to-one with the eigenvectors in U. And λ SThe larger the value, the more overall information the corresponding feature retains. Therefore, after sorting the feature values ​​from largest to smallest, the eigenvectors corresponding to the first j feature values ​​are used to construct the dimensionality reduction matrix Y.

[0150] The formula for constructing the dimensionality reduction matrix is ​​as follows:

[0151] Y = Q r U s×j

[0152] S2.4: Calculate the reconstruction matrix Z and the reconstruction error e. m The reduced-dimensional matrix Y is reconstructed to its original dimension to obtain the reconstructed matrix Z. The formula for calculating the reconstructed matrix is ​​as follows:

[0153]

[0154] The formula for calculating the reconstruction error is as follows:

[0155]

[0156] S2.5: Determine the optimal threshold Ω and make the decision. This algorithm improves the threshold selection method of the PCA anomaly detection algorithm by using the maximization of the objective function method to determine the threshold. Specifically, the reconstruction error is iterated from its minimum to its maximum value, and each time the corresponding reconstruction error is used as the threshold Ω. q Select Ω that maximizes the objective function value. q , which serves as the optimal threshold Ω.

[0157] The objective function formula for this algorithm is as follows:

[0158]

[0159]

[0160]

[0161] Where fp is the false alarm rate, fn is the false negative rate, and M is the total number of training samples. C0 and C1 are the two penalty parameters mentioned in (1.5), and k is a user-selected hyperparameter. In our experiment, k = 1 in the objective function formula.

[0162] After obtaining the optimal threshold Ω, the reconstruction error of the sample to be tested is compared with the optimal threshold Ω. If it is greater than Ω, it is judged as the target; otherwise, it is judged as clutter.

[0163] S3: Input the samples initially identified as clutter in step S2 into a controllable false alarm support vector machine for accurate detection. The controllable false alarm support vector machine used in this step differs from a general support vector machine. Specifically, when determining the decision hyperplane, the quadratic programming problem becomes as follows:

[0164]

[0165] stYi[k(ω,Fi)-b]≥1-ξi,i=1,2...,N

[0166] ξi≥0, i=1,2…,N

[0167] Here, ξi is a slack variable, representing the tolerance for errors. Unlike general support vector machines that use a single penalty parameter, the controllable false alarm support vector machine uses C0 and C1 together to form the penalty parameter, controlling the severity of the penalty for errors. It can be seen that when Yi is +1, the penalty parameter is C1, meaning that in the case of a missed detection, the penalty is controlled by C1. When C is -1, the penalty parameter is C0, meaning that in the case of a false alarm, the penalty is controlled by C0.

[0168] See Figure 2 After the hyperplane is determined, a controllable false alarm support vector machine performs fine-grained detection. Samples classified as targets, along with those initially classified as targets in S2.5, are all output as targets in the final result of the overall detector. Samples classified as clutter are output as clutter in the final result of the overall detector.

[0169] S4: Calculate the actual false alarm rate based on the results obtained in S3 and compare it with the set false alarm rate. If they are not equal, modify C0 and repeat the training process of S2 and S3. If they are equal, proceed to the next step.

[0170] S4.1: Calculate the actual false alarm rate (PF) based on the results obtained in S3 and compare it with the set false alarm rate (PF). If the actual false alarm rate is less than the set false alarm rate, keep C1 unchanged and decrease C0. The formula for decreasing C0 is as follows:

[0171] C h =C0 and

[0172] S4.2: If the actual false alarm rate is greater than the set false alarm rate, keep C1 unchanged and increase C0. The formula for increasing C0 is as follows:

[0173] C l =C0 and

[0174] S4.3: Input the C1 value and the modified C0 value into the improved principal component analysis anomaly detector and controllable false alarm support vector machine, and repeat steps S2 and S3 until the actual false alarm rate is the same as the set false alarm rate. Record the C0 value and C1 value corresponding to the actual false alarm rate being the same as the set false alarm rate.

[0175] S5: Input the C0 and C1 values ​​obtained in S4.3 into the overall detector composed of an improved principal component analysis anomaly detector and a controllable false alarm support vector machine to detect the test sample. This will give the detection result of the test sample at a given false alarm rate.

[0176] Example 3

[0177] Referring to Table 1, an embodiment of the present invention provides an information tree construction method suitable for MMS message analysis. To verify its beneficial effects, the beneficial effects are explained by comparison.

[0178] In the experiment, some classic and recent target detection algorithms in the field of small target detection at sea were selected for comparison. The specific experimental results are shown in the table below.

[0179] Table 1 Comparison of experimental results

[0180]

[0181] The experiments show that, compared to other detectors used in the experiments, the detector proposed in this invention significantly improves the average detection rate across the ten datasets used in the experiments, with improvements ranging from 5% to 30%, across four polarization modes and with false alarm rates of 0.001 and 0.01. The experiments demonstrate that the method of this invention can effectively control the false alarm rate of the detection results regardless of whether the signal-to-noise ratio is high or low. Furthermore, under the same false alarm rate, this algorithm achieves a higher detection rate, effectively improving detection performance.

[0182] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting small targets on the sea surface, characterized in that, include: The radar echo is preprocessed, which includes extracting radar echo signal data and initializing detection algorithm parameters to obtain input data. The input data is fed into an improved principal component analysis-based anomaly detector for preliminary detection to obtain a first clutter sample. The preliminary detection includes determining the optimal threshold by using the objective function maximization method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold. The first clutter sample is input into a controllable false alarm support vector machine for accurate detection to obtain the second clutter sample. The actual false alarm rate is calculated based on the output results of the preliminary detection and the precise detection and compared with the set false alarm rate. The comparison is used to adjust the two penalty parameters. The preliminary detection and precise detection are repeated until the actual false alarm rate is the same as the set false alarm rate, and the optimized penalty parameters are obtained. The optimized penalty parameters are input into the overall detector to detect the samples and obtain the classification results.

2. The method for detecting small targets on the sea surface as described in claim 1, characterized in that, The extracted radar echo signal data includes amplitude, fractal, and frequency information; The initialization detection algorithm parameters include the penalty for false alarms. The severity of penalties for missed detections The upper limit of the penalty for false alarms The lower bound of the change in the penalty for false alarms Compensation parameters for the penalty severity in the anomaly detection phase for false alarms. .

3. The method for detecting small targets on the sea surface as described in claim 1 or 2, characterized in that, The preliminary tests also include, The temporal entropy (TIE), time hearst exponent (THE), and frequency peak-to-average ratio (FPAR) features obtained in the preprocessing are fused into a three-dimensional feature, denoted as: And the training matrix is ​​constructed using the three-dimensional features. , for OK A matrix of columns, where S is the dimension of the feature space of the samples; For the training matrix Standardization yields the matrix and to Singular value decomposition is performed on the covariance matrix, and the eigenvector matrix is ​​obtained after decomposition. and eigenvalue diagonal matrix ; Constructing a dimensionality reduction matrix Determine the eigenvector matrix The feature vectors to be retained are sorted by feature values ​​from largest to smallest. The eigenvectors corresponding to the eigenvalues ​​are used to construct the dimensionality reduction matrix. ; Calculate the reconstruction matrix And calculate the reconstruction error of a single sample. , reduce the dimension of the matrix Reconstructing back to the original dimensions yields the reconstruction matrix. ; The reconstruction error is expressed as follows: in, To reconstruct the matrix, This represents a single sample in the training set. is a single dimension in the feature space dimension of the sample.

4. The method for detecting small targets on the sea surface as described in claim 3, characterized in that, The optimal threshold is determined by maximizing the objective function. The reconstruction error is iterated from its minimum to its maximum value, and the corresponding reconstruction error is used as the threshold at each step. Select the threshold that maximizes the objective function value. As the optimal threshold The objective function is expressed as: in The number of false alarms This represents the number of missed detections. The total number of samples in the training set. and There are two penalty parameters. These are the hyperparameters that you selected yourself.

5. The method for detecting small targets on the sea surface as described in claim 4, characterized in that, The process of obtaining the first clutter sample includes obtaining the optimal threshold. Then, the reconstruction error of the test sample and the optimal threshold are used. The comparison shows that the value is greater than the optimal threshold. If the target is not identified, it is considered a target; otherwise, it is considered clutter.

6. The method for detecting small targets on the sea surface as described in claim 5, characterized in that, The first clutter sample is input into a controllable false alarm support vector machine for accurate detection to obtain a second clutter sample, including determining a decision hyperplane, denoted as: in, As slack variables, Together, they constitute the penalty parameters to control the severity of the penalty for errors. When the value is +1, the penalty parameter is taken as... ;when When the value is -1, the penalty parameter takes... .

7. The method for detecting small targets on the sea surface as described in claim 5 or 6, characterized in that, The actual false alarm rate is calculated based on the output results of preliminary and precise detections and compared with a set false alarm rate. This comparison is used to adjust two penalty parameters, including... If the actual false alarm rate is less than the set false alarm rate, then maintain... Unchanged, decreased , is represented as: If the actual false alarm rate is greater than the set false alarm rate, then maintain... Unchanged, increased , is represented as: 。 8. A system for detecting small targets on the sea surface, using the method described in any one of claims 1-7, characterized in that, include, The preprocessing module is used to preprocess the radar echo, the preprocessing including extracting radar echo signal data and initializing detection algorithm parameters to obtain input data; The preliminary detection module is used to input the input data into an improved principal component analysis-based anomaly detector for preliminary detection to obtain a first clutter sample. The preliminary detection includes determining the optimal threshold by maximizing the objective function method, iterating the reconstruction error from the minimum to the maximum value, using the corresponding reconstruction error as the threshold each time, and selecting the threshold that maximizes the objective function value as the optimal threshold. The precision detection module is used to input the first clutter sample into a controllable false alarm support vector machine for precision detection in order to obtain the second clutter sample; The comparison and judgment module is used to calculate the actual false alarm rate based on the output results of the preliminary detection and the precise detection and compare it with the set false alarm rate. The comparison is used to adjust the two penalty parameters and to cycle through the preliminary detection and precise detection until the actual false alarm rate is the same as the set false alarm rate, so as to obtain the optimized penalty parameters. The results classification module is used to input the optimized penalty parameters into the overall detector to detect samples and obtain classification results.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the sea surface small target detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the sea surface small target detection method according to any one of claims 1 to 7.

Images