A Method and System for Identifying Marine Hybrid Targets Based on Mode Decomposition and Reconstruction

Abstract

This application belongs to the field of radar signal processing and target recognition technology, specifically relating to a method and system for identifying maritime hybrid targets based on mode decomposition and reconstruction. The method includes: performing variational mode decomposition on the radar echo signal of a maritime hybrid target, decomposing the original signal into multiple intrinsic mode signals; estimating the background noise energy benchmark based on the decomposed mode signals and removing noise modes; clustering the remaining modes according to their center frequencies, merging modes with frequencies within the center frequency range into those belonging to the same target, and reconstructing them into independent single-target signals; performing time-frequency analysis on each reconstructed single-target signal to obtain a corresponding time-frequency map, and extracting time-frequency domain features from the time-frequency map; inputting the extracted features into a trained support vector machine classifier to achieve automatic identification of ship targets and floating targets. This method can more accurately distinguish between ship targets and floating targets in hybrid scenarios, improving the stability of tracking and identification.

Description

Technical Field

[0001] This application belongs to the field of radar signal processing and target recognition technology, specifically relating to a method and system for recognizing marine hybrid targets based on mode decomposition and reconstruction. Background Technology

[0002] A mixed target at sea refers to a situation where a ship and a floating target are within the same radar range cell, forming a "mixed" echo. Compared to a single target signal, a mixed echo has greater structural complexity. In the frequency domain, its energy distribution often exhibits a multi-peak structure; in the time-frequency diagram, it manifests as multiple prominent micro-Doppler ridges, further increasing the difficulty of identification.

[0003] Existing floating target identification methods often fail when dealing with mixed-body problems. Current algorithms primarily focus on the Doppler channel with the strongest energy in the time-frequency graph. If the ship's echo energy is higher than that of the floating target, the system tends to identify the mixed-body as a ship. If the floating target's echo energy is dominant, the system may misidentify it as a floating target, missing continuous tracking of the ship or even causing a "tracking failure". If the two energies are close, the system is prone to frequently switching between the two target types, seriously affecting the stability of tracking and identification. Summary of the Invention

[0004] In order to achieve accurate separation and classification of targets in a hybrid body and improve the integrity of system perception, this invention provides a method and system for identifying targets in a marine hybrid body based on modal decomposition and reconstruction.

[0005] In a first aspect, the present invention provides a method for identifying marine hybrid targets based on mode decomposition and reconstruction, comprising the following steps:

[0006] S1. Perform variational mode decomposition on the radar echo signal of the mixed marine target to decompose the original signal into multiple intrinsic mode signals;

[0007] S2. Perform mode filtering on the decomposed modal signals, including:

[0008] Calculate the modal energy peak value for each modal signal obtained from the decomposition;

[0009] Calculate the background noise energy reference of the original signal;

[0010] Based on the background noise energy benchmark, and combined with a preset threshold factor, the energy threshold is dynamically set, and modes with modal energy peak values ​​less than the energy threshold are identified as noise modes and eliminated.

[0011] S3. Cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single target signals;

[0012] S4. Perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features.

[0013] S5. Input the extracted features into the trained support vector machine classifier to achieve automatic identification of ship targets and floating targets.

[0014] Variational Mode Decomposition (VMD) decomposes complex signals into multiple intrinsic mode signals, effectively separating different components in a mixed target and providing a clear signal foundation for subsequent feature extraction and classification. Mode filtering and clustering reconstruction steps eliminate noisy modes and merge modes belonging to the same target, effectively improving signal integrity and signal-to-noise ratio, and enhancing target recognition accuracy. Instantaneous micro-Doppler frequency total variation features and principal Doppler channel entropy features extracted from the time-frequency domain accurately reflect the target's motion characteristics and signal complexity, providing high-quality feature input for the classifier. Inputting the extracted features into a trained support vector machine classifier enables automatic identification of ship targets and floating targets, improving recognition efficiency and accuracy while reducing manual intervention.

[0015] As a further limitation of the technical solution of the present invention, in S2, the step of calculating the modal energy peak value of each modal signal includes:

[0016] The first obtained from mode decomposition k Modal signals Perform amplitude squaring operation;

[0017] Find the maximum value of the square of the amplitude across all time domains; this is the nth... k The peak modal energy of a modal signal; the formula is as follows:

[0018]

[0019] In the formula, For the first k The modal energy peak value of a modal signal, where t is time.

[0020] By using amplitude squaring and maximum value extraction, the modal energy peak value of each modal signal can be accurately calculated, providing a reliable basis for subsequent noise mode removal. The calculation process is simple and easy to implement, enabling rapid evaluation of the energy characteristics of modal signals without increasing computational complexity.

[0021] As a further limitation of the technical solution of the present invention, in S2, the step of calculating the background noise energy reference of the original signal includes:

[0022] Extract the signal segment corresponding to the set frequency from the original signal spectrum as the noise segment. The average energy of the noise segment is calculated as the background noise energy reference; the set frequency is the set percentage frequency of the highest frequency in the original signal spectrum;

[0023]

[0024] In the formula, M is the background noise energy reference, M is the number of sampling points in the noise segment, and t is time.

[0025] By extracting the signal segment corresponding to the set frequency in the original signal spectrum as the noise segment and calculating its average energy as the background noise energy benchmark, this method of setting the frequency can more reasonably select the noise segment, making the calculated background noise energy benchmark more in line with the actual situation, providing an accurate basis for dynamically setting the energy threshold, and enhancing the accuracy of mode screening.

[0026] As a further limitation of the technical solution of the present invention, in S2, the modal signal satisfies the following condition:

[0027]

[0028] If it is determined to be a noise mode, it will be eliminated. Let be the peak modal energy of the i-th modal signal. This is the threshold factor.

[0029] Specific conditions for determining noise modes are provided. Based on the background noise energy benchmark, and combined with a preset threshold factor, the energy threshold is dynamically set. Modes with energy peak values ​​lower than the energy threshold are identified as noise modes and eliminated. This dynamic threshold setting method can adapt to noise conditions under different environments and signal conditions, more accurately eliminate noise modes, reduce the impact of noise on subsequent target recognition, and improve recognition accuracy.

[0030] As a further limitation of the technical solution of the present invention, S3 specifically includes:

[0031] Calculate the center frequency of each remaining modal component;

[0032] A center frequency deviation Δf is preset; all remaining modes are clustered according to their center frequencies, and modal components with a center frequency difference less than Δf are grouped into the same cluster.

[0033] For each cluster, the modal components belonging to that cluster are weighted and summed to obtain the reconstructed single-target signal;

[0034] Perform an inverse Fourier transform on the reconstructed signal obtained from each cluster to obtain the single-target signal in the time domain.

[0035] As a further limitation of the technical solution of the present invention, in S4, the calculation method of the instantaneous micro-Doppler frequency total variation characteristics includes:

[0036] Extract the frequency position sequence of the dominant frequency ridge line from the time-frequency graph;

[0037] The instantaneous micro-Doppler frequency total variation characteristics are obtained by performing first-order difference on the frequency position sequence and summing the absolute values.

[0038] This feature is obtained by extracting the frequency position sequence of the dominant frequency ridge in the time-frequency graph and summing the absolute values ​​of the first-order differences. This feature can well reflect the instantaneous change characteristics of the target signal in the time-frequency domain, which helps to distinguish between ship targets and floating targets, provides more discriminative feature information for the classifier, and improves the recognition accuracy.

[0039] As a further limitation of the technical solution of the present invention, in S4, the method for calculating the main Doppler channel level entropy feature includes:

[0040] Integrate the time-frequency plot along the time direction to obtain the total energy of each Doppler channel;

[0041] The channel with the highest energy is selected as the main Doppler channel, and its energy sequence in the time direction is extracted.

[0042] The hierarchical entropy characteristics of the main Doppler channel are obtained by calculating the hierarchical entropy of the energy sequence.

[0043] First, the total energy of each Doppler channel is obtained by integrating the time-frequency plot along the time direction. The channel with the highest energy is selected as the main Doppler channel. Then, its energy sequence along the time direction is extracted and the hierarchical entropy is calculated. This feature can characterize the complexity of the energy sequence of the main Doppler channel, providing the classifier with important features about the energy of the target signal, and further enhancing the classifier's ability to distinguish between different targets.

[0044] As a further limitation of the technical solution of the present invention, the step of calculating the level entropy characteristics of the main Doppler channel by performing level entropy calculation on the energy sequence includes:

[0045] With a preset embedding dimension m, the energy sequence is reconstructed into a matrix;

[0046] Sort the elements in the matrix to obtain a vector consisting of the number of swaps;

[0047] Calculate the probability of various number of swaps;

[0048] Based on the calculated probability, the entropy value is calculated using Shannon entropy. and ;

[0049] Entropy value and Normalization yields the hierarchical entropy, which in turn gives the hierarchical entropy characteristics of the main Doppler channel.

[0050] By reconstructing a matrix with a preset embedding dimension, sorting the matrix elements to obtain a vector of exchange counts, calculating the probability of various exchange counts, and then calculating the entropy value based on Shannon entropy and normalizing it to obtain the hierarchical entropy. This detailed calculation process ensures the accuracy and consistency of hierarchical entropy feature calculation, providing a guarantee for accurately extracting the features of the target signal and helping to improve the performance of target recognition.

[0051] As a further limitation of the technical solution of the present invention, the energy sequence

[0052] matrix

[0053] A vector consisting of the number of swaps

[0054] Probability of various number of swaps

[0055] Entropy

[0056] Level Entropy

[0057] In the formula, s represents the energy sequence, m represents the embedding dimension, and N is the length of the energy sequence. Indicates when exist The remainder represents the number of occurrences when the values ​​are the same, and c represents the number of swaps. and These are the entropy values ​​for embedding dimensions m and m+1, respectively.

[0058] As a further limitation of the technical solution of the present invention, in S5, the support vector machine classifier is a multinomial kernel function support vector machine classifier, which projects the input features into a high-dimensional space through nonlinear mapping, and describes the high-order correlation between features for automatic identification of ship targets and floating targets.

[0059] Secondly, the technical solution of the present invention also provides a marine hybrid target identification system based on mode decomposition and reconstruction, comprising:

[0060] The mode decomposition module is used to perform variational mode decomposition on the radar echo signal of the mixed marine target, decomposing the original signal into multiple intrinsic mode signals;

[0061] The modal filtering module is used to filter the decomposed modal signals. Specifically, it is used to calculate the modal energy peak of each modal signal; calculate the background noise energy benchmark of the original signal; and dynamically set the energy threshold based on the background noise energy benchmark and a preset threshold factor to identify and remove modes whose modal energy peak is less than the energy threshold as noise modes.

[0062] The clustering and reconstruction module is used to cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single-target signals;

[0063] The time-frequency analysis module is used to perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features.

[0064] The classifier module is used to input the extracted features into a trained support vector machine classifier to achieve automatic identification of ship targets and floating targets.

[0065] As a further limitation of the technical solution of the present invention, the modal screening module includes:

[0066] The modal energy peak unit is used to perform amplitude square operation on the k-th modal signal obtained from modal decomposition; it is used to find the maximum value of the amplitude square in all time domains as the modal energy peak of the k-th modal signal.

[0067] The background noise energy benchmark calculation unit is used to extract the signal segment corresponding to a set frequency in the original signal spectrum as a noise segment, wherein the set frequency is a set percentage frequency of the highest frequency in the original signal spectrum; and calculate the average energy of the noise segment as the background noise energy benchmark.

[0068] The filtering unit is used to dynamically set an energy threshold based on the relationship between the energy peak value of the modal signal and the background noise energy reference, combined with a preset threshold factor, and to identify and remove modal energy peak values ​​that are less than the energy threshold value as noise modes.

[0069] The clustering and reconstruction module is specifically used to calculate the center frequency of each remaining modal component; preset a center frequency deviation Δf; cluster all remaining modes according to their center frequencies, and group modal components with center frequency differences less than Δf into the same cluster; for each cluster, perform a weighted sum of all modal components belonging to that cluster to obtain the reconstructed single-target signal; and perform an inverse Fourier transform on the reconstructed signal obtained from each cluster to obtain the single-target signal in the time domain.

[0070] As a further limitation of the technical solution of the present invention, the time-frequency analysis module includes:

[0071] The time-frequency analysis unit is used to perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram.

[0072] The first calculation unit is used to extract the frequency position sequence of the main frequency ridge in the time-frequency diagram; the frequency position sequence is subjected to first-order difference and the sum of absolute values ​​is obtained to obtain the instantaneous micro-Doppler frequency total variation characteristics.

[0073] The second calculation unit is used to integrate the time-frequency graph along the time direction to obtain the total energy of each Doppler channel, select the channel with the largest energy as the main Doppler channel, and extract its energy sequence in the time direction; perform hierarchical entropy calculation on the energy sequence to obtain the hierarchical entropy characteristics of the main Doppler channel.

[0074] As a further limitation of the technical solution of the present invention, the second calculation unit is specifically used to preset the embedding dimension m, reconstruct the energy sequence into a matrix; sort the elements in the matrix to obtain a vector composed of the number of exchanges; calculate the probability of various exchange numbers; calculate the entropy value through Shannon entropy based on the calculated probability; and normalize the entropy value to obtain the level entropy, that is, obtain the main Doppler channel level entropy feature.

[0075] As a further limitation of the technical solution of the present invention, the classifier module is a multinomial kernel function support vector machine classifier, comprising:

[0076] The feature input unit is used to input the extracted features into a multinomial kernel function support vector machine classifier.

[0077] The nonlinear mapping unit is used to project the input features into a high-dimensional space through nonlinear mapping, thereby characterizing the higher-order relationships between features.

[0078] The classification unit is used to automatically identify ship targets and floating targets.

[0079] As can be seen from the above technical solution, this application has the following advantages: By performing variational mode decomposition on the radar echo signal, the complex hybrid signal is decomposed into multiple intrinsic mode signals, which are then screened and clustered to reconstruct independent single-target signals. Finally, time-frequency analysis and feature extraction are performed, and automatic identification is achieved using a support vector machine classifier. This processing method effectively solves the identification problem caused by the complex structure of hybrid signals, can more accurately distinguish between ship targets and floating targets, and improves the stability of tracking and identification. Attached Figure Description

[0080] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0081] Figure 1 This is a flowchart illustrating the method provided in an embodiment of the present invention.

[0082] Figure 2 The spectrum of the hybrid target.

[0083] Figure 3 This is the spectrum diagram after mode decomposition.

[0084] Figure 4 This is the filtered spectrum.

[0085] Figure 5 To reconstruct the spectrum of the signal.

[0086] Figure 6 This is a system block diagram provided for an embodiment of the present invention. Detailed Implementation

[0087] To achieve accurate separation and classification of targets within a hybrid system, this application proposes a target recognition method based on multi-target mode decomposition reconstruction and time-frequency feature differences. The core idea is to introduce a Variational Mode Decomposition (VMD) algorithm to decompose radar echo signals into multiple modal components, thereby achieving Doppler channel separation. To address the modal redundancy and fragmented target information representation across multiple modes caused by traditional mode decomposition after setting a preset number of modes, an energy-constrained filtering module and a spectrum consistency-based clustering module are designed after VMD processing to achieve automatic filtering and effective integration of modal signals. Specifically, the filtering module approaches the problem from the energy dimension, identifying and eliminating redundant modes with levels close to background noise, thereby suppressing information interference introduced by non-target components and effectively improving the signal-to-noise ratio and feature separability of the reconstructed signal. Furthermore, to address the fragmentation problem caused by the cross-modal distribution of target information, this application designs a clustering module that clusters the decomposition results based on the modal center frequency, identifies modes with adjacent frequencies as originating from the same target, and thus completes modal merging and full reconstruction.

[0088] The preprocessed signal is called the reconstructed signal, and a short-time Fourier transform is performed on the reconstructed signal to obtain its time-frequency information. From an image perspective, the total variation of instantaneous micro-Doppler frequency (VF) is used to measure the fluctuation of the "ridges" or "energy trajectories" in the target micro-Doppler image. From the perspective of data information structure, this application introduces the rating entropy of the dominant doppler channel (REDDC) to measure the disorder of the temporal energy distribution of the signal in the dominant frequency channel. Finally, the feature vector is formed and input into a multinomial kernel SVM to achieve target classification.

[0089] To make the objectives, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solutions protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0090] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0091] like Figure 1 As shown, this embodiment of the invention provides a method for identifying marine hybrid targets based on mode decomposition and reconstruction, including the following steps:

[0092] S1. Perform variational mode decomposition on the radar echo signal of the mixed marine target to decompose the original signal into multiple intrinsic mode signals;

[0093] This application employs existing technologies to acquire radar echo signals from marine hybrid targets. In this step, variational mode decomposition is performed on the radar echo signals, decomposing the original signal into a set of intrinsic mode components (IMFs) with specific frequency characteristics. These specific frequency characteristics refer to the unique frequency distribution characteristics of each IMF signal, including center frequency, bandwidth, frequency distribution uniformity, and frequency characteristic adaptability.

[0094] To ensure effective separation of the components in the signal while preserving modal redundancy space, the modal component number K is set to a value greater than the target number, for example, 4.

[0095] Taking measured data as an example, Figure 2 The spectrum of the hybrid target shows two energy peaks. The energy peak located at zero frequency with a wider bandwidth represents a floating target; the energy peak not located at zero frequency with a narrower bandwidth represents a ship target. Mode decomposition is performed on the hybrid target echo. Figure 3 The spectrum after mode decomposition is shown below. Figure 3 It can be seen that mode 1 corresponds to the ship mode, modes 3 and 4 correspond to the floating target mode, and mode 2 is the background noise mode.

[0096] S2. Modal filtering is performed on the decomposed modal signals. To ensure that the decomposed modes accurately correspond to ships and floating targets, a two-stage processing method is designed. First, modes with energy similar to the background noise are filtered out by energy discrimination to eliminate noise components that do not contain effective target information. Then, modes with similar frequencies are merged based on the center frequency characteristics to ensure that targets with a wide spectrum, such as floating targets, can be presented in complete modal form, thereby improving the signal reconstruction quality and recognition reliability. Figure 3 It can be seen that the energy of background noise mode 2 is almost identical to the noise floor energy. This characteristic, where the peak energy of this mode's spectrum is approximately the same as the noise floor energy, can be used to design a filtering module. This module can filter out irrelevant modes that characterize background noise. This step specifically includes:

[0097] Calculate the modal energy peak value for each modal signal obtained from the decomposition; calculate the background noise energy benchmark of the original signal; based on the background noise energy benchmark, dynamically set the energy threshold in combination with the preset threshold factor, and identify and remove modes whose modal energy peak value is less than the energy threshold as noise modes.

[0098] Since the modal energy of background noise is similar to that of the noise floor, and the center frequencies of the target signals are different, this application designs filtering and clustering modules based on this. The energy peak values ​​of all modal components are calculated, and a signal segment corresponding to 20% of the high frequencies in the original signal spectrum is selected as a noise segment. The average energy of this noise segment is calculated as the noise energy benchmark, and an energy threshold is set based on the noise energy benchmark. Modes with modal energy peak values ​​less than the energy threshold are identified as noise modes and are then filtered out.

[0099] First, extract the signal segment corresponding to the highest 20% of the frequency spectrum from the original signal and denote it as the noise segment. Calculate the average energy of this noise segment as a baseline for background noise energy:

[0100]

[0101] Where M is the number of sampling points in the noise segment, and t is time. Subsequently, for the th... k Modal signals Calculate its peak energy:

[0102]

[0103] If the mode satisfies the following condition:

[0104]

[0105] If it is determined to be a noise mode, it will be eliminated. Let be the peak modal energy of the i-th modal signal. It is a threshold factor that is adaptively and dynamically adjusted based on the relative amplitude of the highest modal peak and the noise reference. Figure 3 The spectrum diagram after mode decomposition and Figure 4 The filtered spectrum was compared. Figure 4 As can be seen, mode 2, which characterizes background noise, has been successfully filtered out.

[0106] S3. Cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single target signals;

[0107] By merging modes with similar frequencies based on the center frequency characteristics, targets with a wide frequency spectrum, such as floating targets, can be presented in complete mode form, thereby improving the signal reconstruction quality and recognition reliability.

[0108] In cases where the center frequencies of ship targets and floating targets are far apart (if the ship and floating targets overlap in the distance cell and in the frequency dimension, it is not within the scope of this application), and the two modes of a single target decomposition have relatively close center frequencies in the spectrum, this application introduces a clustering module to cluster modes with similar center frequencies and merge them into a complete mode. Figure 4 This is the filtered spectrum, which is actually the spectrum before modal clustering. Figure 5 To reconstruct the spectrum of the signal. Figure 4 and Figure 5 Comparison shows that this method effectively solves the fragmentation problem of floating target signals being split into multiple modes during mode decomposition, thereby improving the integrity of the reconstructed signal.

[0109] In summary, for hybrid targets, this application effectively separates the hybrid echo signal into multiple independent single-target signals by combining mode decomposition with a frequency domain energy filtering module and a clustering module.

[0110] S4. Perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features.

[0111] In order to quantify the structural differences between ship targets and floating targets in the time and frequency domain, this application extracts two types of micro-motion features from the image structure level and the time series data level, respectively, to characterize the motion regularity and frequency stability of the targets.

[0112] Step a: Extract the instantaneous micro-Doppler frequency total variation features;

[0113] I: First, find the frequency position sequence of the dominant frequency ridge line in the time-frequency graph:

[0114]

[0115] This is a time-frequency graph matrix, with the vertical axis representing frequency f and the horizontal axis representing time t, respectively. Let t be the frequency value at time t, and T be the time length.

[0116] II: Perform first-order differencing on the frequency position sequence and sum the absolute values ​​to obtain the total temporal variation of the dominant frequency ridge:

[0117]

[0118] The frequency value at time t-1.

[0119] Step b: Extract the level entropy features of the main Doppler channel;

[0120] I: First, extract data from the main Doppler channel and generate the time-frequency graph matrix. Integrating along the time direction, we obtain the total energy of each Doppler channel:

[0121]

[0122] II: Select the channel with the highest energy. Using the main Doppler channel, extract the energy sequence of this channel in the time direction:

[0123]

[0124] III: Obtaining the main Doppler channel energy sequence After N is the length of the sequence and the preset embedding dimension m, the reconstructed matrix is... It can be represented as:

[0125]

[0126] IV: For Each element in the array is sorted in ascending order according to bubble sort, resulting in a vector consisting of the number of swaps. :

[0127]

[0128] V: By normalization, find the probabilities of various exchanges. ,in Indicates when exist The number of times the values ​​in the middle are the same.

[0129]

[0130] Where c represents the number of swaps, ranging from 0 to... .

[0131] VI: The probability and entropy value obtained based on the above formula. It can be obtained through Shannon entropy:

[0132]

[0133] VII: Set m to m+1 and repeat to get the result. , will get and Normalization yields the level entropy:

[0134]

[0135] S5. Input the extracted features into the trained support vector machine classifier to achieve automatic identification of ship targets and floating targets.

[0136] In target classification tasks, traditional linear SVMs have limitations in discriminative ability due to the complex nonlinear relationships between features. To address this, this application introduces a multinomial kernel function, projecting the original data into a high-dimensional space through nonlinear mapping, thereby characterizing higher-order relationships between features. The decision boundary constructed within this space can more effectively distinguish different categories, exhibiting stronger robustness, especially in low signal-to-noise ratio environments, significantly improving classification performance.

[0137] It should be noted that the classifier training process includes: combining the extracted instantaneous micro-Doppler frequency total variation (VF) and main Doppler channel entropy (REDDC) features to form a feature vector for classification. A multinomial kernel function is chosen as the kernel function of the support vector machine (SVM). This kernel function maps the feature vector from the original low-dimensional space to a high-dimensional space to capture the nonlinear relationship between features. The constructed SVM model is trained using labeled sample data containing ship targets and floating targets to determine the optimal parameters of the model, enabling it to accurately distinguish between the two types of targets. The feature vector extracted from the reconstructed signal to be identified is input into the trained SVM model. The model makes a classification judgment based on the distribution of the feature vector in the high-dimensional space and outputs the target category (ship target or floating target) corresponding to the feature vector, thereby achieving automatic identification.

[0138] The proposed method was validated using experimental data. Hybrid target datasets and independent target datasets were selected to verify the performance of the proposed method.

[0139] Several sets of mixed-target and single-target data in multi-target scenarios were selected to verify the performance of the proposed algorithm in detecting the existence of mixed targets. Specific data descriptions are shown in Table 1, which includes three sets of mixed-target data and one set of independent target data.

[0140] Table 1: Dataset Description

[0141]

[0142] To evaluate the algorithm's performance in detecting the presence of mixed targets, detection was performed on mixed targets included in datasets 1, 2, and 3. If the algorithm determined a target to be a mixed target, it was considered a correct detection. For dataset 4, independent ship targets and independent buoy targets were selected for detection. If the algorithm determined a target to be a non-mixed target, it was considered a correct detection, used to evaluate the algorithm's false alarm rate.

[0143] Table 2: Performance Analysis of Hybrid Target Presence Detection

[0144]

[0145] As shown in Table 2, in datasets 1 and 2, the mixed targets are located in sea state 2. While the detection accuracy decreases with shorter observation times, the average accuracy remains at 94.96%. In dataset 3, the mixed targets are located in sea state 4, and the detection accuracy is significantly improved compared to sea state 2. For dataset 4, which only contains independent ship and buoy targets and no mixed targets, the results show an extremely low false alarm rate of only 0.1% when facing non-mixed targets, indicating a good false alarm suppression effect.

[0146] In this embodiment of the invention, step S2, calculating the modal energy peak value of each modal signal, includes:

[0147] The k-th mode signal obtained from mode decomposition Perform amplitude squaring operation;

[0148] Find the maximum value of the square of the amplitude across all time domains; this is the nth... k The modal energy peak value of a modal signal.

[0149] In some embodiments, S3 specifically includes:

[0150] Calculate the center frequency of each remaining modal component;

[0151] A center frequency deviation Δf is preset; all remaining modes are clustered according to their center frequencies, and modal components with a center frequency difference less than Δf are grouped into the same cluster.

[0152] For each cluster, the modal components belonging to that cluster are weighted and summed to obtain the reconstructed single-target signal;

[0153] Perform an inverse Fourier transform on the reconstructed signal obtained from each cluster to obtain the single-target signal in the time domain.

[0154] like Figure 6 As shown, this embodiment of the invention also provides a marine hybrid target identification system based on mode decomposition and reconstruction, comprising:

[0155] The mode decomposition module is used to perform variational mode decomposition on the radar echo signal of the mixed marine target, decomposing the original signal into multiple intrinsic mode signals;

[0156] The modal filtering module is used to filter the decomposed modal signals. Specifically, it is used to calculate the modal energy peak of each modal signal; calculate the background noise energy benchmark of the original signal; and dynamically set the energy threshold based on the background noise energy benchmark and a preset threshold factor to identify and remove modes whose modal energy peak is less than the energy threshold as noise modes.

[0157] The clustering and reconstruction module is used to cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single-target signals;

[0158] The time-frequency analysis module is used to perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features.

[0159] The classifier module is used to input the extracted features into a trained support vector machine classifier to achieve automatic identification of ship targets and floating targets.

[0160] The entire system can automatically complete the entire process from signal input to target recognition, effectively solving the problem of target recognition in mixed marine environments and improving the stability and accuracy of maritime target tracking and recognition.

[0161] In some embodiments, the modality screening module includes:

[0162] The modal energy peak unit is used to perform amplitude square operation on the k-th modal signal obtained from modal decomposition; it is used to find the maximum value of the amplitude square in all time domains as the modal energy peak of the k-th modal signal.

[0163] The background noise energy benchmark calculation unit is used to extract the signal segment corresponding to a set frequency in the original signal spectrum as a noise segment, wherein the set frequency is a set percentage frequency of the highest frequency in the original signal spectrum; and calculate the average energy of the noise segment as the background noise energy benchmark.

[0164] The filtering unit is used to dynamically set an energy threshold based on the relationship between the energy peak value of the modal signal and the background noise energy reference, combined with a preset threshold factor, and to identify and remove modal energy peak values ​​that are less than the energy threshold value as noise modes.

[0165] The clustering and reconstruction module is specifically used to calculate the center frequency of each remaining modal component; preset a center frequency deviation Δf; cluster all remaining modes according to their center frequencies, and group modal components with center frequency differences less than Δf into the same cluster; for each cluster, perform a weighted sum of all modal components belonging to that cluster to obtain the reconstructed single-target signal; and perform an inverse Fourier transform on the reconstructed signal obtained from each cluster to obtain the single-target signal in the time domain.

[0166] In some embodiments, the time-frequency analysis module includes:

[0167] The time-frequency analysis unit is used to perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram.

[0168] The first calculation unit is used to extract the frequency position sequence of the main frequency ridge in the time-frequency diagram; the frequency position sequence is subjected to first-order difference and the sum of absolute values ​​is obtained to obtain the instantaneous micro-Doppler frequency total variation characteristics.

[0169] The second calculation unit is used to integrate the time-frequency graph along the time direction to obtain the total energy of each Doppler channel, select the channel with the largest energy as the main Doppler channel, and extract its energy sequence in the time direction; perform hierarchical entropy calculation on the energy sequence to obtain the hierarchical entropy characteristics of the main Doppler channel.

[0170] In some embodiments, the second computing unit is specifically used to preset the embedding dimension m, reconstruct the energy sequence into a matrix; sort the elements in the matrix to obtain a vector composed of the number of exchanges; calculate the probability of various exchange numbers; calculate the entropy value through Shannon entropy based on the calculated probability; and normalize the entropy value to obtain the level entropy, that is, obtain the main Doppler channel level entropy feature.

[0171] In some embodiments, the classifier module is a multinomial kernel function support vector machine classifier, including:

[0172] The feature input unit is used to input the extracted features into a multinomial kernel function support vector machine classifier.

[0173] The nonlinear mapping unit is used to project the input features into a high-dimensional space through nonlinear mapping, thereby characterizing the higher-order relationships between features.

[0174] The classification unit is used to automatically identify ship targets and floating targets.

[0175] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying marine hybrid targets based on mode decomposition and reconstruction, characterized in that, Includes the following steps: S1. Perform variational mode decomposition on the radar echo signal of the mixed marine target to decompose the original signal into multiple intrinsic mode signals; S2. Perform mode filtering on the decomposed modal signals, including: Calculate the modal energy peak value for each modal signal obtained from the decomposition; Calculate the background noise energy reference of the original signal; Based on the background noise energy benchmark, and combined with a preset threshold factor, the energy threshold is dynamically set, and modes with modal energy peak values ​​less than the energy threshold are identified as noise modes and eliminated. S3. Cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single target signals; S4. Perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features. S5. Input the extracted features into the trained support vector machine classifier to achieve automatic identification of ship targets and floating targets; The calculation method for the instantaneous micro-Doppler frequency total variation characteristics includes: Extract the frequency position sequence of the dominant frequency ridge line from the time-frequency graph; By performing first-order differences on the frequency position sequence and summing the absolute values, the instantaneous micro-Doppler frequency total variation characteristics are obtained. The calculation method for the main Doppler channel level entropy characteristics includes: Integrate the time-frequency plot along the time direction to obtain the total energy of each Doppler channel; The channel with the highest energy is selected as the main Doppler channel, and its energy sequence in the time direction is extracted. The hierarchical entropy characteristics of the main Doppler channel are obtained by calculating the hierarchical entropy of the energy sequence; specifically including: With a preset embedding dimension m, the energy sequence Reconstructed into a matrix ; Sort the elements in the matrix to obtain a vector consisting of the number of swaps. ; Calculate the probability of various number of swaps. ; Based on the calculated probability, the entropy value is calculated using Shannon entropy. ; Increase the embedding dimension m to m+1, and calculate the result in the same way. ; to entropy value and Normalization yields the hierarchical entropy, which in turn yields the hierarchical entropy characteristics of the main Doppler channel; In the formula, s represents the energy sequence, and N is the length of the energy sequence. Indicates when exist The value in the middle represents the number of occurrences when the values ​​are the same, and c represents the number of swaps.

2. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 1, characterized in that, In S2, the steps for calculating the modal energy peak value of each modal signal include: The first obtained from mode decomposition k Modal signals Perform amplitude squaring operation; Find the maximum value of the square of the amplitude across all time domains; this is the nth... k The peak modal energy of a modal signal; the formula is as follows: In the formula, Let t be the peak modal energy of the k-th modal signal, and t be time.

3. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 2, characterized in that, In S2, the steps for calculating the background noise energy reference of the original signal include: Extract the signal segment corresponding to the set frequency from the original signal spectrum as the noise segment. The average energy of the noise segment is calculated as the background noise energy reference; the set frequency is the set percentage frequency of the highest frequency in the original signal spectrum; In the formula, M is the background noise energy reference, M is the number of sampling points in the noise segment, and t is time.

4. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 3, characterized in that, In S2, the modal signals satisfy the following conditions: If it is determined to be a noise mode, it will be eliminated. Let be the peak modal energy of the i-th modal signal. This is the threshold factor.

5. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 4, characterized in that, S3 specifically includes: Calculate the center frequency of each remaining modal component; A center frequency deviation Δf is preset; all remaining modes are clustered according to their center frequencies, and modal components with a center frequency difference less than Δf are grouped into the same cluster. For each cluster, the modal components belonging to that cluster are weighted and summed to obtain the reconstructed single-target signal; Perform an inverse Fourier transform on the reconstructed signal obtained from each cluster to obtain the single-target signal in the time domain.

6. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 5, characterized in that, In the step of calculating the hierarchical entropy of the energy sequence to obtain the hierarchical entropy characteristics of the main Doppler channel, the matrix... : Entropy : Level Entropy In the formula, and These are the entropy values ​​for embedding dimensions m and m+1, respectively.

7. The method for identifying marine hybrid targets based on mode decomposition and reconstruction according to claim 6, characterized in that, In S4, the support vector machine classifier is a multinomial kernel function support vector machine classifier, which projects the input features into a high-dimensional space through nonlinear mapping, characterizing the high-order correlation between features for automatic identification of ship targets and floating targets.

8. A marine hybrid target identification system based on mode decomposition and reconstruction for implementing the method of claim 1, characterized in that, include: The mode decomposition module is used to perform variational mode decomposition on the radar echo signal of the mixed marine target, decomposing the original signal into multiple intrinsic mode signals; The modal filtering module is used to filter the decomposed modal signals, specifically to calculate the modal energy peak of each modal signal and to calculate the background noise energy benchmark of the original signal. Based on the background noise energy benchmark, and combined with a preset threshold factor, the energy threshold is dynamically set, and modes with modal energy peak values ​​less than the energy threshold are identified as noise modes and eliminated. The clustering and reconstruction module is used to cluster the remaining modes according to the center frequency, merge the modes with frequencies within the center frequency range into the same target, and reconstruct them into independent single-target signals; The time-frequency analysis module is used to perform time-frequency analysis on each reconstructed single target signal to obtain the corresponding time-frequency diagram, and extract time-frequency domain features from the time-frequency diagram, including instantaneous micro-Doppler frequency total variation features and main Doppler channel level entropy features. The classifier module is used to input the extracted features into a trained support vector machine classifier to achieve automatic identification of ship targets and floating targets.

Images