A Sea Surface Small Target Detection Method Based on Four FAST Features

Abstract

The present invention discloses a method for detecting small sea targets based on four FAST features in the field of radar signal processing, comprising the following steps: First, obtain the signal to be detected and perform denoising preprocessing on the signal to be detected; obtain the required normalized time-frequency distribution spectrogram; use the FAST algorithm to extract four features and perform normalization processing, and combine the four normalized features into a four-dimensional feature vector; obtain the optimal hyperparameter group and update the decision threshold; input the data sample into the classifier to obtain a predicted value, and judge whether there is a target in the echo signal by comparing with the decision threshold. The method of the present invention can greatly accelerate the operation speed of the network; effectively improve the detection probability and enhance the network detection performance; when the observation duration is 0.512 s, the detection probability of the detection method of the present invention is increased by 7%, and when the observation duration is 1.024 s, the detection probability of the detection method proposed by the present invention is increased by 13.8%, and it can be applied to the detection of small sea targets.

Description

Technical Field

[0001] The present invention belongs to the field of radar signal processing, and particularly relates to a method for detecting small sea targets based on four FAST features. Background Technique

[0002] China not only has a vast land area, but also has a very vast sea area and a long coastline. In recent years, with the continuous improvement of the value of the ocean, studying the characteristics of sea clutter and the detection technology of small sea targets under the background of sea clutter has extremely important strategic significance. Affected by factors such as the small Radar Cross Section (RCS) of small sea targets and complex sea conditions, their signals are very weak, easily submerged under the background of sea clutter, and difficult to detect, resulting in a low detection rate. Therefore, how to find the differences between suitable sea clutter and target signals is an important research topic.

[0003] Currently, the methods for detecting small targets under the background of sea clutter mainly fall into two categories. The first category analyzes the characteristics of sea clutter from the perspective of statistical models. Under the background of sea clutter, the most widely used statistical model is the K-distribution model, which takes into account the correlation of radar echo data and can better fit the real distribution. However, the sea surface is affected by multiple factors such as wind and tides and has very strong non-stationarity, while the statistical characteristic models of sea clutter are often established under the condition of a stable sea level. Therefore, in the case of complex sea conditions and low signal-to-clutter ratio, the detection effect of statistical models is poor. The second category is based on non-linear methods, directly analyzing the properties of sea clutter from the spatio-temporal perspective. In recent years, the method of neural network learning has also developed to a certain extent in the detection of sea clutter targets, and the detection method of small sea targets with new multi-domain and multi-dimensional features is still concerned by the academic community. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method for detecting small sea targets based on four FAST features to solve the problems raised in the above background technique.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for detecting small sea targets based on four FAST features includes the following steps:

[0007] Step 1: First, obtain the signal to be detected and perform denoising preprocessing on the signal to be detected. The signal to be detected includes sea clutter signals and sea clutter and target echo signals. Labels "0" and "1" are respectively assigned to the sea clutter signals and target echo signals, and the detection problem is reduced to a binary hypothesis test;

[0008] Step 2: Transform the original signal into a two-dimensional time-frequency distribution spectrogram through short-time Fourier transform, obtain the energy distribution information of the original signal through modulus operation and perform normalization processing to obtain the required normalized time-frequency distribution spectrogram;

[0009] Step 3: Use the FAST algorithm to extract four features and perform normalization processing, and combine the four normalized features into a four-dimensional feature vector;

[0010] Step 4: Construct an XGBoost classifier model to evaluate and classify the extracted features, use the genetic optimization algorithm to optimize the XGBoost network hyperparameter group, obtain the optimal hyperparameter group and update the decision threshold;

[0011] Step 5: Input the data sample into the classifier to obtain a predicted value, and judge whether there is a target in the echo signal by comparing with the decision threshold.

[0012] Preferably, the binary hypothesis test in Step 1 is as follows:

[0013]

[0014] In the formula, c represents the sea clutter signal, s represents the target echo signal, the H0 hypothesis means that there is only the sea clutter signal in the echo signal, and the H1 hypothesis means that there is a target echo in the echo signal.

[0015] Preferably, Step 2 specifically includes the following steps:

[0016] Step 2.1: Perform short-time Fourier transform on the original signal, and the formula is as follows:

[0017]

[0018] Step 2.2: Perform modulus operation, and the operation formula is:

[0019] SPEC(t,ω) = ||STFT(t,ω)||

[0020] Step 2.3: Perform normalization processing on the time-frequency distribution spectrogram, and the normalization formula is:

[0021]

[0022] Step 2.4: During the operation process, the average function μ(n,l) and the standard deviation function σ(n,l) are involved, and their formulas are:

[0023]

[0024]

[0025] Preferably, the four FAST features extracted in step 3 are as follows:

[0026] ξ1: The number of candidate feature points in the time-frequency distribution spectrogram;

[0027] ξ2: The average distribution energy of candidate feature points;

[0028] ξ3: The number of clusters obtained by clustering candidate feature points;

[0029] ξ4: The number of candidate feature points in the largest cluster.

[0030] Preferably, the formula for normalizing the four features in step 3 is as follows:

[0031]

[0032] Preferably, each feature is taken as a dimension to construct a four-dimensional feature space, and a four-dimensional space vector is obtained:

[0033]

[0034] Preferably, step 4 specifically includes the following steps:

[0035] Step 4.1: Set the range of hyperparameters in the XGBoost network and perform binary encoding on the hyperparameter group;

[0036] Step 4.2: Calculate the fitness value of the hyperparameter group, and the fitness value is selected as the detection probability of the signal to be detected by the XGBoost network;

[0037] Step 4.3: Use the selection operator, crossover operator, and mutation operator to iteratively update the hyperparameters by the genetic algorithm;

[0038] Step 4.4: Determine whether the fitness value converges. If the fitness value converges, the current iteration terminates and the optimal hyperparameter group is output; otherwise, return to step 4.3 and continue the iteration;

[0039] Step 4.5: Obtain n predicted values of the actually sea clutter data evaluated and classified by the detector, and arrange them from largest to smallest, denoted as ρ1, ρ2,..., ρ n , select the false alarm rate P fa , and its calculation formula is

[0040]

[0041] Step 4.6: Calculate the decision threshold γ, and its calculation formula is

[0042]

[0043] Preferably, the judgment criterion in step 5 is as follows:

[0044] When ρ > γ, it is determined that there is a target in the detection signal, belonging to the H1 hypothesis; when ρ < γ, it is determined that there is no target in the detection signal, belonging to the H0 hypothesis.

[0045] Advantages of the present invention:

[0046] 1. The method of the present invention uses the short-time Fourier transform to map the original signal into a two-dimensional time-frequency distribution spectrogram, effectively extracting the energy distribution information of the signal. Due to the convenient and fast characteristics of FAST, using the FAST algorithm to extract four features can greatly accelerate the operation speed of the network. At the same time, using the FAST algorithm can effectively extract corner features, thereby increasing the difference between the sea clutter and the target echo features;

[0047] 2. The method of the present invention uses the XGBoost algorithm as the classifier network, and at the same time uses the genetic optimization algorithm to optimize the hyperparameter group of the XGBoost algorithm, effectively improving the detection probability and enhancing the network detection performance;

[0048] 3. When the observation duration is 0.512 s, the detection probability of the detection method of the present invention is increased by 7%. When the observation duration is 1.024 s, the detection probability of the detection method proposed by the present invention is increased by 13.8%, which can be applied to the detection of small targets on the sea surface. Description of the Drawings

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0050] Figure 1 It is a schematic flowchart of the method of the present invention;

[0051] Figure 2 It is a comparison diagram of the distribution of candidate feature points of sea clutter and target echo data under HH polarization in the present invention;

[0052] Figure 3 It is a comparison diagram of the detection performance of the FAST four-feature detection method under four polarization modes in the present invention;

[0053] Figure 4 It is a comparison diagram of the detection performance of five detection methods under four polarization modes in the present invention (N = 512, P fa = 10 -3 );

[0054] Figure 5It is a comparison chart of the detection performance of five detection methods under four polarization modes in the present invention (N = 1024, P fa = 10 -3 ). Specific embodiments

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Please refer to Figure 1 As shown, the present invention proposes a sea surface small target detection method based on four FAST features, including the following steps:

[0057] Step 1: First, obtain the signal to be detected and perform denoising preprocessing on the signal to be detected. The signal to be detected includes sea clutter signal and sea clutter and target echo signal, and the "0" label and "1" label are respectively assigned to the sea clutter signal and the target echo signal;

[0058] The detection problem is reduced to a binary hypothesis test:

[0059]

[0060] Among them, c represents the sea clutter signal, s represents the target echo signal, the H0 hypothesis represents that there is only the sea clutter signal in the echo signal, and the H1 hypothesis represents that the echo signal contains the target echo.

[0061] Step 2: Convert the original signal into a two-dimensional time-frequency distribution spectrogram through short-time Fourier transform (STFT), obtain the energy distribution information of the original signal through modulus operation, and perform normalization processing;

[0062] It includes the following steps:

[0063] Step 2.1: Perform short-time Fourier transform on the original signal, and the formula is as follows:

[0064]

[0065] In the formula, h(τ) is the window function, and s(t) is the original signal;

[0066] Step 2.2: Perform modulus operation, and the operation formula is:

[0067] SPEC(t,ω) = ||STFT(t,ω)||

[0068] Step 2.3: Perform normalization processing on the time-frequency distribution spectrogram, and the normalization formula is:

[0069]

[0070] Step 2.4. During the operation process, the average function μ(n, l) and the standard deviation function σ(n, l) are involved, and their formulas are as follows:

[0071]

[0072]

[0073] Step 2.5. Obtain the required normalized time-frequency distribution spectrogram through the above steps.

[0074] Step 3. Use the FAST algorithm to extract four features and perform normalization processing, and combine the four normalized features into a four-dimensional feature vector;

[0075] It includes the following steps:

[0076] Step 3.1. Use the FAST algorithm to extract four features. The four FAST features extracted are as follows:

[0077] ξ1: The number of candidate feature points in the time-frequency distribution spectrogram;

[0078] ξ2: The average distribution energy of candidate feature points;

[0079] ξ3: The number of clusters obtained by clustering candidate feature points;

[0080] ξ4: The number of candidate feature points in the largest cluster.

[0081] In the formula, the eigenvalue ξ1 is directly obtained by the FAST algorithm by counting candidate feature points. The eigenvalues ξ2 and ξ3 are calculated from the eigenvalue ξ1. The DBSCAN algorithm is used to calculate ξ3. For ξ4, the cluster with the most candidate feature points is selected from all the clusters of ξ3, and then the number of candidate feature points within the cluster is counted;

[0082] Step 3.2. When extracting the feature ξ3, the clustering method DBSCAN algorithm used, by setting the maximum scan radius r max and the minimum number of points contained Pt min , starting from any one candidate feature point, if the number of points within the scan radius is greater than Pt min , then a cluster is formed with the current point and nearby points; if the number of points within the scan radius is less than Pt min , it is considered a noise point and is removed;

[0083] Step 3.3. Perform normalization processing on the four features, and its formula is:

[0084]

[0085] Step 3.4: Taking each feature as a dimension, construct a four-dimensional feature space to obtain a four-dimensional space vector:

[0086]

[0087] Step 4: Construct an XGBoost classifier model to evaluate and classify the extracted features. Use a genetic optimization algorithm to optimize the XGBoost network hyperparameter group, obtain the optimal hyperparameter group, and update the decision threshold;

[0088] It includes the following steps:

[0089] Step 4.1: Set the range of hyperparameters in the XGBoost network and perform binary encoding on the hyperparameter group;

[0090] Step 4.2: Calculate the fitness value of the hyperparameter group. The fitness value is selected as the detection probability of the signal to be detected by the XGBoost network;

[0091] Step 4.3: Use the selection operator, crossover operator, and mutation operator to perform iterative updates on the hyperparameters using the genetic algorithm. The three operators are as follows:

[0092] (1) Selection operator: Calculate the fitness value of each individual through the roulette wheel method and perform random selection based on the roulette wheel ratio;

[0093] (2) Crossover operator: Set a threshold k and perform crossover at the k-th position of the gene;

[0094] (3) Mutation operator: Select the gene mutation position according to the mutation probability and perform 0-1 conversion to prevent local optimization;

[0095] Step 4.4: Determine whether the fitness value converges. If the fitness value converges, terminate the current iteration and output the optimal hyperparameter group; otherwise, return to Step 4.3 and continue the iteration.

[0096] Step 4.5: Obtain n predicted values of the actually sea clutter data evaluated and classified by the detector, and arrange them from largest to smallest, denoted as ρ1, ρ2, …, ρ n , select the false alarm rate P fa , and its calculation formula is

[0097]

[0098] Step 4.6: Calculate the decision threshold γ, and its calculation formula is

[0099]

[0100] Step 5: Input the data sample into the classifier to obtain the predicted value. By comparing it with the decision threshold, determine whether there is a target in the echo signal. The judgment criterion is as follows: when ρ > γ, it is judged that there is a target in the detection signal, belonging to the H1 hypothesis; when ρ < γ, it is judged that there is no target in the detection signal, belonging to the H0 hypothesis.

[0101] The data used in this paper is from the IPIX radar target database, which was collected by Professor Haykin of McMaster University in Canada on the east coast of Canada. 10 groups of data were taken in the experiment. Each group of data consists of 14 adjacent range cells. Each range cell contains 131072 pulses. The range resolution is 30m. The target is a polystyrene foam ball wrapped with wire mesh, with a diameter of about 1m. Four polarization modes are obtained according to different data transmission and reception methods, namely HH, HV, VH, and VV.

[0102] Figure 2 It is a comparison chart of candidate feature points extracted by FAST in sea clutter and echo with target. Select the data of the 1st cell gate (sea clutter) and the 8th cell gate (echo with target) of #54 data. Set the observation time to 0.512s, that is, each group of data contains N = 512 points. The data of the two cell gates are respectively divided into 256 groups, a total of 512 groups. As Figure 2 It is a distribution chart of candidate feature points screened from the fifth group of data of the 1st cell gate and the 8th cell gate. The red dots in the figure are the screened candidate feature points. It can be seen from the figure that there are more candidate feature points screened from the sea clutter data and they are widely distributed, while there are very few candidate feature points screened from the echo data with target. This is because the time-frequency distribution spectrogram of sea clutter is rougher than that of the echo with target, and it is easier to detect corner points that can be used as candidate feature points. In addition, the candidate feature points screened from the sea clutter data are more discrete, which also leads to more clusters after clustering of the candidate feature points screened from the sea clutter data than those of the echo with target, and the number of candidate feature points in the largest cluster is also more. Thus, it can be seen that the features ξ1, ξ2, and ξ4 are somewhat distinguishable to a certain extent.

[0103] Figure 3 It is a comparison chart of the detection performance of the FAST four-feature detection method under four polarization modes. Among them, the false alarm rate P fa = 10 -3 , N = 512 in (a) and N = 1024 in (b). It can be seen that the detection performance under the HH, HV, and VH polarization modes is good, and at the same time, the detection probabilities of HV and VH are better than those of HH and VV.

[0104] Figure 4 、 Figure 5Performance comparison diagrams of four detectors under four polarizations for 10 groups of data respectively under N = 512 and N = 1024. It can be seen that the detection performance of the detection method proposed by the present invention is generally better than the other three detection methods. Among them, when N = 512, the detection method proposed by the present invention is improved by 7%; when N = 1024, the detection method proposed by the present invention is improved by 13.8%. It can be seen that the detection method proposed by the present invention has better detection effect.

[0105] In summary, the present invention solves the problems of difficult design and complex calculation of the target detection method for feature extraction, and effectively improves the detection probability of small floating targets on the sea surface.

[0106] In the description of this specification, the descriptions with reference to the terms "an embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0107] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0109] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device, and the instruction device implements in the processFigure 1 one process or multiple processes and / or boxes Figure 1 the functions specified in one box or multiple boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or boxes Figure 1 one box or multiple boxes.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific implementation manners of the present invention. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for detecting small sea surface targets based on four FAST features, characterized in that, It includes the following steps: Step 1: First, obtain the signal to be detected and perform denoising preprocessing on the signal to be detected. The signal to be detected includes sea clutter signals and sea clutter and target echo signals. "0" labels and "1" labels are respectively assigned to the sea clutter signals and target echo signals, and the detection problem is reduced to a binary hypothesis test; Step 2: Transform the original signal into a two-dimensional time-frequency distribution spectrogram through short-time Fourier transform, obtain the energy distribution information of the original signal through modulus operation and perform normalization processing to obtain the required normalized time-frequency distribution spectrogram; Step 3: Use the FAST algorithm to extract four features and perform normalization processing, and combine the four normalized features into a four-dimensional feature vector; Step 4: Construct an XGBoost classifier model, evaluate and classify the extracted features, use the genetic optimization algorithm to optimize the XGBoost network hyperparameter group, obtain the optimal hyperparameter group and update the decision threshold; Step 5: Input the data sample into the classifier to obtain a predicted value, and judge whether there is a target in the echo signal by comparing with the decision threshold.

2. The sea surface small target detection method based on four FAST features according to claim 1, characterized in that The binary hypothesis test in Step 1 is as follows: In the formula, c represents the sea clutter signal, s represents the target echo signal, the H0 hypothesis represents that there is only the sea clutter signal in the echo signal, and the H1 hypothesis represents that there is a target echo in the echo signal.

3. A method for detecting small sea surface targets based on four FAST features according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Perform short-time Fourier transform on the original signal, and the formula is as follows: In the formula, h(τ) is the window function, and s(t) is the original signal; Step 2.2: Perform modulus operation, and the operation formula is: SPEC(t,ω) = ||STFT(t,ω)|| Step 2.3: Perform normalization processing on the time-frequency distribution spectrogram, and the normalization formula is: In the operation process, the average function μ(n,l) and the standard deviation function σ(n,l) are involved, and their formulas are:

4. A method for detecting small sea targets based on four FAST features according to claim 1, characterized in that The four FAST features extracted in Step 3 are as follows: ξ1: The number of candidate feature points in the time-frequency distribution spectrogram; ξ2: The average distribution energy of the candidate feature points; ξ3: The number of clusters obtained by clustering the candidate feature points; ξ4: The number of candidate feature points in the largest cluster; In the formula, the eigenvalue ξ1 is directly obtained by counting the candidate feature points by the FAST algorithm, the eigenvalues ξ2 and ξ3 are calculated from the eigenvalue ξ1, the DBSCAN algorithm is used to calculate ξ3, and ξ4 selects the cluster with the most candidate feature points among all the clusters of ξ3, and then counts the number of candidate feature points in the cluster.

5. The method for detecting small sea targets based on four FAST features according to claim 4, wherein When extracting the feature ξ3, the DBSCAN algorithm, a clustering method, is used. By setting the maximum scanning radius r max and the minimum number of points Pt min , starting from an arbitrary candidate feature point, if the number of points within the scanning radius is greater than Pt min , a cluster is formed with the current point and nearby points; if the number of points within the scanning radius is less than Pt min , it is considered a noise point and removed.

6. The method for detecting small sea targets based on four FAST features according to claim 5, characterized in that The formula for normalizing the four features in Step 3 is as follows:

7. A method for detecting small sea targets based on four FAST features according to claim 6, characterized in that, Taking each feature as a dimension, construct a four-dimensional feature space to obtain a four-dimensional space vector:

8. A method for detecting small sea targets based on four FAST features according to claim 1, characterized in that Step 4 specifically includes the following steps: Step 4.1: Set the range of hyperparameters in the XGBoost network and perform binary coding processing on the hyperparameter group; Step 4.2: Calculate the fitness value of the hyperparameter group, and the fitness value is selected as the detection probability of the signal to be detected by the XGBoost network; Step 4.3: Use the selection operator, crossover operator, and mutation operator to perform genetic algorithm iterative update on the hyperparameters. Step 4.4: Determine whether the fitness value converges. If the fitness value converges, the current iteration terminates, and the optimal hyperparameter set is output; Otherwise, return to Step 4.3 and continue the iteration; Step 4.5: Obtain n sets of predicted values that are actually sea clutter data evaluated and classified by the detector, and arrange them in descending order, denoted as ρ1, ρ2, …, ρ n , and select the false alarm rate P fa , and its calculation formula is Step 4.6: Calculate the decision threshold γ, and its calculation formula is 9. A method for detecting small sea targets based on four FAST features according to claim 8, characterized in that, The selection operator is to calculate the fitness value of each individual through the roulette method to form the roulette ratio for random selection; the crossover operator is to set a threshold k and perform crossover at the k-th position of the gene; the mutation operator is to select the gene mutation position according to the mutation probability and perform 0-1 conversion to prevent local optimization.

10. A method for detecting small sea targets based on four FAST features according to claim 9, characterized in that, The judgment criterion in Step 5 is as follows: When ρ > γ, it is judged that there is a target in the detection signal, belonging to the H1 hypothesis; when ρ < γ, it is judged that there is no target in the detection signal, belonging to the H0 hypothesis.

Images