A method for identifying sea surface targets in radar staring mode

Abstract

The present invention relates to a method for identifying sea surface targets in the radar staring mode, belonging to the technical field of radar signal processing. To solve the deficiency that the accuracy of the existing radar target recognition method based on the one-dimensional range profile features of the target is low, the method includes the following steps: Step 1, echo data acquisition to obtain the range cell where the target is located; Step 2, time-domain processing and time-domain feature extraction of the echo data according to the range cell where the target is located obtained in Step 1; Step 3, frequency-domain feature extraction of the echo data; Step 4, constructing a feature sequence; Step 5, optimizing and training a classifier: optimizing the parameters of the SVM classifier and training to obtain an optimal classifier model; Step 6, target recognition: inputting the data to be recognized into the classifier model trained in Step 5 to realize the recognition of ship targets and floating targets. This method has strong applicability and high accuracy.

Description

Technical Field

[0001] The present invention relates to a method for identifying sea surface targets, specifically a method for identifying sea surface targets in the radar staring mode, and belongs to the technical field of radar signal processing. Background Art

[0002] When the radar detects the sea, accurately identifying and classifying sea surface targets and determining their specific categories, uses, and threat levels are of great significance in both civilian and military fields. Common sea surface floating targets include small fishing boats, buoys, ice floes, etc. Traditional radar target recognition methods are based on the one-dimensional range profile features of targets. When the radar beam irradiates along the ship's side, such targets have similar one-dimensional range profile characteristics to ship targets such as large cruise ships and cargo ships, seriously affecting the recognition performance. In addition, when the radar resolution decreases or the signal-to-clutter ratio (SCR) of the target is low, the accuracy of the recognition method relying on the amplitude features of the target drops significantly. Summary of the Invention

[0003] The purpose of the present invention is to solve the deficiency of the low accuracy of the existing radar target recognition method based on the one-dimensional range profile features of targets, and provide a method for identifying sea surface targets in the radar staring mode, which uses the periodic consistency of the transient power level and bandwidth change of the target to distinguish floating targets from ship targets.

[0004] Starting from the coupling mechanism between the target and the sea surface, exploring the differences between ship targets and sea surface floating targets in the staring mode, and constructing a statistic that can describe the differences between the two can effectively solve the above problems. Obviously, the mass distribution of floating targets is uniform and the density is small. Under staring conditions, the coupling degree with sea waves is high, and the periodicity of fluctuating up and down with sea waves is obvious; ship targets are large in size and complex in structure, and the distribution of scattering points is unstable due to their own movement or uneven structure, and the coupling degree with sea waves is low, and the periodicity of parameters is not obvious. Therefore, the corresponding relationship between the transient power where sea clutter exists and the periodicity of the bandwidth at the centroid is more obvious in sea surface floating targets.

[0005] To solve the above problems, the present application is implemented through the following technical solutions:

[0006] A method for identifying sea surface targets in the radar staring mode, which is characterized in that it includes the following steps:

[0007] Step 1, collecting echo data to obtain the range cell where the target is located;

[0008] Step 2, performing time-domain processing and time-domain feature extraction on the echo data according to the range cell where the target is located obtained in Step 1;

[0009] Step 3, Frequency-domain feature extraction of echo data: In each data segment, calculate the Fourier transform (FFT) of each segment of data to obtain the Doppler spectrum of this segment of data. Use a non-parametric method to obtain the centroid sequence of the corresponding Doppler spectrum in each data segment and the root-mean-square bandwidth sequence BW at the centroid;

[0010] Step 4, Constructing the feature sequence: Assign label values to each feature vector. Normalize the transient echo power level sequence obtained in Step 2 and the root-mean-square bandwidth sequence at the centroid obtained in Step 3 to obtain a feature sequence, and divide the feature sequence into a training set and a test set;

[0011] Step 5, Optimizing and training the classifier: Optimize the parameters of the SVM classifier and perform training to obtain an optimal classifier model;

[0012] Step 6, Target recognition: Input the data to be recognized into the classifier model trained in Step 5 to achieve the recognition of ship targets and floating targets.

[0013] Furthermore, the specific content of Step 1 is as follows: The sea detection radar scans and detects sea surface targets. After detecting a target, it conducts a staring observation of the target at a fixed angle. The radar receiver receives the target echo signal. The echo data set in the staring mode is denoted as ,

[0014] where: is the number of frames of radar-acquired data, , represents the time-domain echo data of dimension in the th frame of data, represents the number of pulses included in each frame, represents the number of range cells in the staring scene;

[0015] For the case where range cell migration occurs for moving targets, use non-coherent integration frame by frame to align the range cells: , , is the range cell where the target is located in the th frame of data.

[0016] Furthermore, the specific content of Step 2 is as follows:

[0017] Set the number of pulses included in each data segment and the overlap rate of adjacent data segments, select the range cell where the target is located, and intercept the radar echo along the time dimension ;

[0018] In each data segment Within, perform a sliding window process on the in-segment data, remove the speckle component, and calculate the transient echo power level sequence of this segment of data: ,

[0019] Where: is the complex value of the th sampling point, represents the transient echo power level sequence of the th data segment,

[0020] Furthermore, the specific calculation method of step 3 is as follows:

[0021] Step 3.1: Calculate the spectral function of each segment of data:

[0022] Select an appropriate number of FFT points for each segment of data, calculate the short-time power spectrum, and obtain the spectral function of each segment of data: ,

[0023] Where: represents the short-time power spectrum of the th data segment. To eliminate the influence of range side lobes on frequency energy, windowing processing needs to be performed in the time domain, represents the time-domain window function, represents the th data segment of radar echo data, exp represents the exponential function with the natural constant e as the base, n represents the sampling point, T represents the pulse repetition frequency of the radar, represents the serial number of the frequency point, represents the th frequency point;

[0024] Step 3.2: Calculate the Doppler centroid sequence and the bandwidth sequence at the centroid:

[0025] Adopt the non-parametric method in modern signal spectral analysis to obtain the Doppler centroid sequence of the target and the root mean square bandwidth sequence at the centroid: , ,

[0026] Where: represents the power level of the short-time Doppler spectrum of the th data segment, that is: ;

[0027] represents the length of a single data segment, represents the frequency, represents the short-time Doppler spectrum, represents the th data segment of the short-time power spectrum.

[0028] Further, step 4 specifically includes the following steps:

[0029] Step 4.1: Normalize the transient echo power level: ,

[0030] In the formula: represents the maximum value operation, represents the column vector of the target normalized time-domain echo transient power level sequence;

[0031] Step 4.2: Normalize the root mean square bandwidth sequence at the centroid: ,

[0032] In the formula: represents the maximum value operation, represents the column vector composed of the bandwidth sequence at the centroid after target normalization;

[0033] Step 4.3: Combine the column vector of the time-domain echo transient power level sequence normalized in step 4.1 and the column vector of the bandwidth sequence at the centroid normalized in step 4.2 to form a feature sequence. The ratio of the data lengths of the training set and the test set is 3:1. The training set is expressed as: .

[0034] Further, step 5 is specifically as follows:

[0035] Use the data in the training set to optimize the hyperparameters in the SVM classifier by means of iterative loops, including the penalty factor c and the RBF kernel function parameter gamma. The ranges of both parameters start from the minimum value of 2^-8 and end at the maximum value of 2^8. The loop step size is incremented exponentially by one. For the model under each set of parameters, perform five-fold cross-validation using the test set in step 4, select the parameters under the optimal recognition effect, and obtain the classifier model.

[0036] The sea surface target recognition method of this application can utilize radar historical data to improve the utilization rate of target echo information; it is not dependent on the radar detection angle and has strong applicability; the recognition accuracy is high: at the second-level observation time, the recognition accuracy of this method for targets with large quality differences such as ship targets and floating targets can reach more than 90%. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Att Figure 1 is the implementation flowchart of the present invention;

[0038] Att Figure 2 is the transient power level sequence of the ship target and the bandwidth sequence at the centroid proposed by the present invention;

[0039] Appended Figure 3 are the transient power level sequence of the sea surface floating target and the bandwidth sequence at the centroid proposed by the present invention. Specific Embodiments

[0040] The following refers to the accompanying drawings to give the specific embodiments of the present invention to further illustrate the composition of the present invention.

[0041] Embodiment 1. A method for identifying sea surface targets in the radar staring mode, the specific process is as Figure 1 shown, including the following steps:

[0042] Step 1, echo data acquisition, obtaining the range cell where the target is located;

[0043] Step 2, perform time-domain processing and time-domain feature extraction on the echo data according to the range cell where the target is located obtained in Step 1;

[0044] Step 3, frequency-domain feature extraction of echo data: within each data segment, calculate the Fourier transform FFT of each segment of data to obtain the Doppler spectrum of the segment of data, and use a non-parametric method to obtain the centroid sequence of the corresponding Doppler spectrum in each data segment and the root mean square bandwidth sequence BW at the centroid;

[0045] Step 4, construct a feature sequence: assign a label value to each feature vector, 1 for a ship target and 2 for a floating target, perform normalization processing on the transient echo power level sequence obtained in Step 2 and the root mean square bandwidth sequence at the centroid obtained in Step 3 to obtain a feature sequence, and divide the feature sequence into a training set and a test set;

[0046] Step 5, optimize and train the classifier: optimize the parameters of the SVM classifier and perform training to obtain an optimal classifier model;

[0047] Step 6, target recognition: input the data to be recognized into the classifier model trained in Step 5 to realize the recognition of ship targets and floating targets.

[0048] Furthermore, the specific content of Step 1 is as follows: the sea detection radar scans and detects the sea surface target, after detecting the target, stares at the target at a fixed angle, the radar receiver receives the target echo signal, and the echo data set in the staring mode is denoted as ,

[0049] where: is the number of frames of radar-acquired data, , represents the th frame of data and the time-domain echo data with dimension in the Indicates the number of pulses contained in each frame, Indicates the number of range cells in the staring scene;

[0050] For the case where range cell migration occurs for a moving target, non-coherent accumulation is used frame by frame to align the range cells: , , That is, the range cell where the target is located in the frame data.

[0051] Further, the specific steps of step 2 are as follows:

[0052] Set the number of pulses contained in each data segment and the overlap rate of adjacent data segments, select the range cell where the target is located , and intercept the radar echo along the time dimension ;

[0053] In each data segment , perform a sliding window process on the data within the segment to remove the speckle component and calculate the transient echo power level sequence of this segment of data: ,

[0054] In the formula: is the complex value of the th sampling point, represents the transient power level sequence of the echo of the th data segment,

[0055] Further, the specific calculation method of step 3 is as follows:

[0056] Step 3.1: Calculate the spectral function of each data segment modulo:

[0057] Select an appropriate number of FFT points for each data segment, calculate the short-time power spectrum, and obtain the spectral function of each data segment: ,

[0058] In the formula: represents the short-time power spectrum of the th data segment. In order to eliminate the influence of range side lobes on the frequency energy, windowing processing needs to be performed in the time domain, represents the time domain window function, represents the radar echo data of the th data segment, exp represents the exponential function with the natural constant e as the base, n represents the sampling point, T represents the pulse repetition frequency of the radar, represents the serial number of the frequency point, represents the th frequency point;

[0059] Step 3.2: Calculate the Doppler centroid sequence and the bandwidth sequence at the centroid:

[0060] Use the non-parametric method in modern signal spectrum analysis to obtain the Doppler centroid sequence of the target and the root mean square bandwidth sequence at the centroid : , ,

[0061] In the formula: represents the power level of the short-time Doppler spectrum of the th data segment, that is: ;

[0062] represents the length of a single data segment, represents the frequency, represents the short-time Doppler spectrum, represents the th short-time power spectrum of the data segment.

[0063] Furthermore, step 4 specifically includes the following steps:

[0064] Step 4.1: Normalize the transient echo power level: ,

[0065] In the formula: represents the maximum value operation, represents the column vector of the target normalized transient echo power level sequence in the time domain;

[0066] Step 4.2: Normalize the root mean square bandwidth sequence at the centroid : ,

[0067] In the formula: represents the maximum value operation, represents the column vector composed of the target normalized bandwidth sequence at the centroid;

[0068] The transient power level sequence of the ship target and the bandwidth sequence at the centroid are shown in Figure 2 ;

[0069] The transient power level sequence of the sea surface floating target and the bandwidth sequence at the centroid are shown in Figure 3 ;

[0070] Step 4.3: Combine the column vector of the transient echo power level sequence in the time domain normalized in step 4.1 Construct a feature sequence, and the ratio of the data lengths of the training set and the test set is 3:1. The training set is represented as: .

[0071] Further, step 5 is specifically as follows:

[0072] Use the data in the training set to optimize the hyperparameters in the SVM classifier by means of cyclic iteration, including the penalty factor c and the RBF kernel function parameter gamma. The ranges of both parameters start from the minimum value of 2^-8 and end at the maximum value of 2^8. The loop step size is increased exponentially by one. For the model under each set of parameters, perform five-fold cross-validation using the test set in step 4, and select the parameters with the best recognition effect to obtain the classifier model.

[0073] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art of the present invention can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, but will not deviate from the spirit of the present invention or exceed the defined scope.

Claims

1. A method for identifying sea surface targets in radar staring mode, characterized in that: It includes the following steps: Step 1: Echo data acquisition to obtain the range bin where the target is located; Step 2: Perform time-domain processing and time-domain feature extraction on the echo data according to the range bin where the target is located obtained in Step 1; Step 3: Frequency-domain feature extraction of echo data: In each data segment, calculate the Fourier transform FFT of each segment of data to obtain the Doppler spectrum of this segment of data, and use a non-parametric method to obtain the centroid sequence of the corresponding Doppler spectrum in each data segment and the root mean square bandwidth sequence BW at the centroid; Step 4: Construct a feature sequence: Assign a label value to each feature vector, perform normalization processing on the transient echo power level sequence obtained in Step 2 and the root mean square bandwidth sequence at the centroid obtained in Step 3 to obtain a feature sequence, and divide the feature sequence into a training set and a test set; The specific steps of Step 4 include the following: Step 4.1: Normalize the transient echo power level: , In the formula: represents the operation of taking the maximum value; represents the column vector of the target normalized time-domain echo transient power level sequence; Step 4.

2. Normalize the root mean square bandwidth sequence at the centroid as follows: , In the formula: represents the maximum value operation, represents the column vector formed by the bandwidth sequence at the centroid after target normalization; Step 4.3: The column vector of the time-domain echo transient power level sequence after normalization in Step 4.1 and the column vector formed by the bandwidth sequence at the centroid after normalization in Step 4.2 constitute a feature sequence. The ratio of the data lengths of the training set and the test set is 3:

1. The training set is expressed as: ; Step 5: Optimize and train the classifier: Optimize the parameters of the SVM classifier and train to obtain an optimal classifier model; Step 6: Target recognition: Input the data to be recognized into the classifier model trained in Step 5 to realize the recognition of ship targets and floating targets.

2. The method for identifying sea surface targets in the radar staring mode according to claim 1, wherein: The specific steps of Step 1 are as follows: The sea detection radar scans and detects sea surface targets. After detecting a target, it conducts a staring observation on the target at a fixed angle. The radar receiver receives the target echo signal, and the echo data set in the staring mode is denoted as , Wherein: is the number of frames of radar-acquired data, , represents the -dimensional time-domain echo data in the -th frame of data, represents the number of pulses included in each frame, represents the number of range cells in the staring scenario; For the case where range cell migration occurs for a moving target, non-coherent accumulation is used frame by frame to align the range cells: , , That is, the range cell where the target is located in the frame data.

3. A method for identifying sea surface targets in a radar staring mode according to claim 1 or 2, characterized in that: The specific steps of Step 2 are as follows: Set the number of pulses included in each data segment and the overlapping rate of adjacent data segments, and select the range cell where the target is located , intercept the radar echo along the time dimension ; In each data segment In the segment, the data in the segment is processed by sliding window to remove the speckle component and calculate the transient echo power level sequence of the segment data: , where: For the The complex value of the sampling points, express The transient power level sequence of the echo in the data segment, Indicates the length of a single data segment.

4. The method for identifying sea surface targets in the radar staring mode according to claim 3, wherein: The specific calculation method of Step 3 is as follows: Step 3.1: Calculate the spectral function of each segment of data modulo; Select appropriate FFT points for each segment of data, calculate the short-time power spectrum, and obtain the spectral function of each segment of data: , Wherein: represents the short-time power spectrum of the th data segment. In order to eliminate the influence of range side lobes on frequency energy, windowing processing needs to be performed in the time domain. represents the time-domain window function, represents the th radar echo data segment, exp represents the exponential function with the natural constant e as the base, n represents the sampling point, T represents the pulse repetition frequency of the radar, represents the serial number of the frequency point, represents the th frequency point; Step 3.2: Calculate the Doppler centroid sequence and the bandwidth sequence at the centroid; The non-parametric method in modern signal spectrum analysis is used to obtain the Doppler centroid sequence of the target and the root mean square bandwidth sequence at the centroid : , , In the formula: represents the power level of the short-time Doppler spectrum of the th data segment, that is: ; Indicates the length of a single data segment, Indicates the frequency, Indicates the short-time Doppler spectrum, Indicates the short-time power spectrum of the nth data segment.

5. The method for identifying sea surface targets in the radar staring mode according to claim 3, wherein: The specific steps of Step 5 are as follows: Use the data in the training set and use a cyclic iteration method to optimize the hyperparameters in the SVM classifier, including the penalty factor c and the RBF kernel function parameter gamma. The ranges of both parameters start from the minimum value of 2^-8 and end at the maximum value of 2^8. The loop step size is incremented exponentially. For the model under each set of parameters, perform five-fold cross-validation using the test set in Step 4, select the parameters under the optimal recognition effect, and obtain the classifier model.

Images