A rapid detection method for marine radar targets

Abstract

The present invention discloses a method for rapid detection of marine radar targets. Radar echo data samples on a single scan line are selected from offline radar data, including radar echo samples containing targets and pure sea clutter radar echo samples without targets; characteristic parameters are selected, and the echo eigenvalue of the samples is input into an RBF neural network for training to obtain a trained RBF neural network classifier model; the echo data to be measured on a single scan line in the radar image is obtained, and the echo eigenvalue of the echo data to be measured is calculated according to the characteristic parameters to obtain an input feature matrix; the input feature matrix is input into the trained RBF neural network classifier model to obtain a decision function value y; y is compared with the classifier threshold to determine whether there is a target on the scan line: if y is greater than the threshold, there is a target; if y is less than the threshold, there is no target. The present invention can reduce the detection time and improve the target detection efficiency and target detection accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine remote sensing under marine environmental conditions, and relates to a method for rapid detection of marine radar targets, in particular to a method for rapid detection of marine radar targets based on a radial basis function neural network. Background Art

[0002] In a complex marine environment, marine radars are widely used in ship navigation to guide ships to avoid marine obstacles and ensure the navigation safety of ships. Therefore, it is of great significance for marine radars to be able to detect targets quickly and effectively.

[0003] In classical object detection, the constant false alarm rate (CFAR) technique that detects points one by one is mostly adopted. In 2002, Li Lan et al. achieved CFAR object detection for maritime targets. When the waveform of sea clutter changes greatly, the CFAR technique has strong detection performance for detecting targets such as ships at sea, but this method is not applicable to object detection under other sea conditions (Li Lan, Deng Feng, Peng Hailiang. Constant False Alarm Rate Object Detection of Synthetic Aperture Radar Images [J]. Journal of Test and Measurement Technology, 2002(01): 9-13). In 2007, Xu Hui et al. found through experimental verification of a large number of radar images that the dual-threshold CFAR technique can improve the performance of object detection in different angular domains (Xu Hui, Guo Qijun, Liu Hongwei. Application of Dual-Threshold Detection in Wideband Radar Constant False Alarm Rate Detection [J]. Radar Science and Technology, 2007(05): 371-374). In 2015, Tan Jie et al. proposed a dual-threshold CFAR algorithm based on the characteristics of civilian navigation radar (JRC). It can not only achieve radar object detection under different sea condition backgrounds, but also improve the detection performance of the classifier. However, this method increases the complexity of the CFAR algorithm, resulting in low detection efficiency (Tan Jie, Xiong Wei, Zhou Wei. Dual-Threshold Detection Method Based on Constant False Alarm Rate [J]. Radar Science and Technology, 2015, 13(02): 154-158). In 2016, Zhao Likai specifically analyzed the specific performance changes of the classical CFAR classifier in object detection when there are many targets at the edge of relatively uniform sea clutter (Zhao Likai. Research on Radar Target Constant False Alarm Rate Detection Algorithm [D]. Beijing Institute of Technology, 2016.). In 2017, Xu Jianqing et al. proposed an improved CFAR detection algorithm with higher detection efficiency than the traditional dual-parameter CFAR algorithm for object detection in SAR images, but there is still a relatively large false alarm rate (Xu Jianqing, Li Junbao, Ma Yuntong, Peng Yu. SAR Image Object Detection Algorithm Based on Improved CFAR [J]. Navigation, Positioning and Timing, 2017, 4(04): 74-81). In 2019, researchers in Science Letter studied the detection technology for targets with slow moving speeds in synthetic aperture radar images (Radar and Sonar Research; New Data from Naval University of Engineering Illuminate Findings in Radar and Sonar Research (Constant False Alarm Rate Detection of Slow Targets In Polarimetric Along-track Interferometric Synthetic Aperture Radar Imagery) [J]. Science Letter, 2019).In summary, the target detection based on CFAR technology generally has a long detection time. At the same time, the algorithms of the CFAR type are becoming increasingly complex and diverse, resulting in low target detection efficiency. Summary of the Invention

[0004] Aiming at the above-mentioned existing technologies, the technical problem to be solved by the present invention is to provide a fast detection method for marine radar targets based on a radial basis function neural network, solve the influence of strong and complex sea clutter signals on target detection, quickly distinguish pure sea clutter and target echoes, and more efficiently improve the detection performance of the classifier, thereby improving the strong target detection efficiency.

[0005] To solve the above technical problems, a fast detection method for marine radar targets of the present invention includes the following steps:

[0006] Step 1: Select radar echo data samples on a single scan line from the offline radar data. The samples include samples containing target radar echoes and samples of pure sea clutter radar echoes without targets; select characteristic parameters, and input the echo characteristic values of the samples into the RBF neural network for training to obtain a trained RBF neural network classifier model;

[0007] Step 2: Obtain the echo data to be measured on a single scan line in the radar image, calculate the echo characteristic values of the echo data to be measured according to the characteristic parameters selected in Step 1 to obtain an input characteristic matrix;

[0008] Step 3: Input the input characteristic matrix obtained in Step 2 into the trained RBF neural network classifier model to obtain a decision function value y;

[0009] Step 4: Compare the obtained decision function value y with the classifier threshold to determine whether there is a target on the scan line: if y is greater than the threshold, there is a target; if y is less than the threshold, there is no target.

[0010] Further, the characteristic parameters include: mean value, variance and fitting correlation coefficient.

[0011] Further, inputting the echo characteristic values of the samples into the RBF neural network for training to obtain a trained RBF neural network classifier model includes:

[0012] Step 1.1: Initialize the connection weight vector from the hidden layer to the output layer:

[0013] W m =[w m1 ,w m2 ,…w mq T ,(m = 1, 2,…, q)

[0014] ​Wherein, q is the number of hidden layer units, and m represents the m-th layer of the hidden layer;

[0015] The weight initialization formula from the hidden layer to the output layer is:

[0016]

[0017] Wherein, mink is the minimum value of the result predicted by the output of the k-th neuron in the training samples; maxk is the maximum value of the result predicted by the output of the k-th neuron in the training samples, and j represents the j-th neuron in the hidden layer;

[0018] Step 1.2: Initialize the central parameter vector of each neuron in the hidden layer:

[0019] C j =[c j1 ,c j2 ,…c jn T

[0020] Wherein, j represents the j-th neuron in the hidden layer, and n represents the number of sampling points of a single scan line;

[0021] The initial value of the central parameter of the RBF neural network is:

[0022]

[0023] Wherein, mini is the minimum value of all input information of the i-th feature in the training set, maxi is the maximum value of all input information of the i-th feature in the training set, and p represents the total number of neurons in the hidden layer;

[0024] Step 1.3: Initialize the width vector formula:

[0025] D j =[d j1 ,d j2 ,…d jn T

[0026] Wherein, j represents the j-th neuron in the hidden layer, and n represents the number of sampling points of a single scan line;

[0027] Width vector:

[0028]

[0029] Wherein, d f is the width adjustment coefficient, x ij represents the i-th feature value of the j-th scan line, and c ij represents the central parameter value;

[0030] The output calculation formula of the j-th hidden layer neuron:​​

[0031]

[0032] Where X represents the input feature vector, C represents its central parameter vector, and D represents the width vector;

[0033] Step 1.4: The input radar image feature data is:

[0034]

[0035] Calculate the radial basis function as the node activation function:

[0036]

[0037] Where R is the value of the hidden layer neuron;

[0038] Step 1.5: Calculate the neuron output:

[0039]

[0040] Where y k is a specific output element in the output vector Y, and the output vector is:

[0041] Y = [y1, y2,... y q T

[0042] The expected output vector is a zero vector, specifically:

[0043] O = [o1, o2,... o q T

[0044] Where q represents the number of hidden layer units;

[0045] Step 1.6: The RBF neural network adaptively adjusts the connection weight vector, central parameter vector, and width vector of the RBF neural network to the optimal values using the gradient descent method; the calculation formulas for the connection weight vector, central parameter vector, and width vector are respectively:

[0046]

[0047]

[0048]

[0049] Where W represents the training weight vector, C represents its central parameter vector, and D represents the width vector;

[0050] ​​Step 1.7: After the center, width, and weight are updated, according to the given values of η and α and the value of the iteration termination accuracy ε, the calculation formula for the root mean square error RMS of the network output is as follows:

[0051]

[0052] In the formula, O represents the output expectation matrix, N is the number of scan lines. If RMS ≤ ε, the training ends; otherwise, return to Step 1.6 to perform iterative calculations on the connection weight vector, center parameter vector, and width vector.

[0053] Furthermore, the radar is an X-band marine radar.

[0054] Advantages of the present invention: Aiming at the problems of long detection time and low detection efficiency that generally exist in the radar image target detection methods applicable to pulsed X-band marine radars in the prior art, the present invention discloses a fast marine radar target detection method based on a radial basis function neural network. This method performs overall detection on the radial data of each scan line that makes up the complete radar image, and can quickly screen out the scan lines of the radar image with targets. This method only needs to use the sample training set to determine the RBF neural network detector model, and can perform target detection on the single-scan-line data of the on-site radar image. The present invention first proposes a fast target detection algorithm based on the RBF neural network, which can reduce the detection time and greatly improve the target detection efficiency and target detection accuracy. Description of the Drawings

[0055] Figure 1 is the image for removing co-frequency interference;

[0056] Figure 2 is the sample containing target echoes;

[0057] Figure 3 is the pure sea clutter sample;

[0058] Figure 4 is the radial basis neuron model;

[0059] Figure 5 is the data curve graph of a certain scan line;

[0060] Figure 6 is the implementation flowchart;

[0061] Figure 7 is the on-site RBF neural network target detection result graph. Detailed Implementation Modes

[0062] The present invention will be further described below in conjunction with the specification drawings and embodiments.

[0063] Combined with Figure 6 , the present invention includes the following steps:

[0064] Step 1, obtain the RBF neural network classifier model:

[0065] Conduct experiments offline. First, perform co-frequency interference suppression processing on the selected original radar images. For the preprocessed radar images, conduct a preliminary screening. Through the offline radar data, select two types of radar echo data samples on a single scan line. One type is the sample containing target radar echoes, and the other type is the sample of pure sea clutter radar echoes without targets. Select appropriate characteristic parameters and use the RBF neural network for training to obtain the classifier model.

[0066] The radial echo characteristic parameters selected in Step 1 include the mean, variance, and fitting correlation coefficient, etc.

[0067] Using the RBF neural network for training in Step 1 to obtain the classifier model includes the following:

[0068] Step 1.1, initialize the connection weight vector from the hidden layer to the output layer:

[0069] W m =[w m1 ,w m2 ,…w mq T ,(m = 1, 2, …, q)

[0070] In the formula, q is the number of hidden layer units. m represents the m-th layer of the hidden layer.

[0071] Based on the method of center initialization, give the weight initialization formula from the hidden layer to the output layer:

[0072]

[0073] In the formula, mink is the minimum value of the prediction result of the k-th output neuron in the training samples; maxk is the maximum value of the prediction result of the k-th output neuron in the training samples, and j represents the j-th neuron in the hidden layer.

[0074] Step 1.2, initialize the center parameter vector of each neuron in the hidden layer:

[0075] C j =[c j1 ,c j2 ,…c jn T

[0076] In the formula, j represents the j-th neuron in the hidden layer, and n represents the number of sampling points on a single scan line.

[0077] The initial value of the center parameter of the RBF neural network is:

[0078] ​

[0079] Where mini is the minimum value of all input information of the i-th feature in the training set, and maxi is the maximum value of all input information of the i-th feature in the training set. p represents the total number of neurons in the hidden layer.

[0080] Step 1.3, initialize the width vector formula:

[0081] D j = [d j1 , d j2 , … d jn T

[0082] Where j represents the j-th neuron in the hidden layer, and n represents the number of sampling points of a single scan line.

[0083] Width vector:

[0084]

[0085] Where d f is the width adjustment coefficient, x ij represents the i-th feature value of the j-th scan line, and c ij represents the center parameter value.

[0086] Output calculation formula of the j-th hidden layer neuron:

[0087]

[0088] Where X represents the input feature vector, C represents its center parameter vector, and D represents the width vector.

[0089] Step 1.4: The input radar image feature data is:

[0090]

[0091] Calculate the radial basis function as the node activation function, and its formula is as follows:

[0092]

[0093] Where R is the value of the hidden layer neuron.

[0094] Step 1.5, finally calculate the result formula of the neuron output from the above formula:

[0095]

[0096] Where y k is a specific output element in the output vector Y. Among them, the output vector and the expected output vector are respectively: ​

[0097] Y = [y1, y2, … y q T

[0098] O = [o1, o2, … o q T

[0099] where q represents the number of hidden layer units.

[0100] Step 1.6, The RBF neural network uses the gradient descent method to adaptively adjust to the optimal value through learning. The iterative calculation formulas for adjusting the weight parameters, centers, and widths are as follows:

[0101]

[0102]

[0103]

[0104] where W represents the training weight vector, C represents its center parameter vector, and D represents the width vector.

[0105] Step 1.7, After updating the centers, widths, and weights, according to the given values of η and α and the iterative termination accuracy ε value, the calculation formula for the root mean square error RMS of the network output is:

[0106]

[0107] where O represents the output expectation matrix and N is the number of scan lines. If RMS ≤ ε, the training ends; otherwise, according to Step 1.6, the weight, center, and width parameters are iteratively calculated until the condition RMS ≤ ε is satisfied.

[0108] Step 2, Obtain the echo data to be measured on a single scan line in the radar image, and select the radar echo data feature parameters according to Step 1 to obtain the input feature matrix.

[0109] Step 3, Input the feature matrix calculated in Step 2 into the RBF classifier model obtained in Step 1 to finally obtain the output value.

[0110] Step 4, Determine whether there is a target:

[0111] Compare the obtained decision function value y with the threshold of the classifier to determine whether there is a target on this scan line. If y is greater than the threshold, there is a target; if the output is less than the threshold, there is no target.

[0112] The following gives an embodiment in combination with specific parameters:

[0113] ​​The marine radar used in the embodiments of the present invention is an X-band navigation radar, which operates in a short-pulse mode. The time for the radar antenna to scan one week is about 2.5 s. The total number of lines in a radar image is 2048, and there are 2048 pixel points on each line. Its azimuth resolution is about 0.1°, and the radial resolution is about 2.5 m.

[0114] The main technical parameters of the marine radar are shown in Table 1:

[0115] Table 1 Technical parameters of the marine radar

[0116]

[0117] Combined with Figure 6 , the specific experimental steps of the present invention are as follows:

[0118] Step 1: Obtain the RBF neural network classifier model.

[0119] Step 1.1: Conduct off-line experiments. First, perform co-frequency interference suppression processing on the selected original radar image through median filtering. Conduct preliminary screening on the preprocessed radar image, such as Figure 1 . In this embodiment, two types of radar echo data samples on 5559 scan lines of radar data are selected for training. One type is 2704 samples containing target radar echoes, such as Figure 2 . The other type is 2855 samples of pure sea clutter radar echoes without targets, such as Figure 3 .

[0120] Step 1.2: In this example, three types of parameters such as the mean, variance, and fitting correlation coefficient of the radar echo sample data are selected as feature parameters and their eigenvalues are calculated. The calculations of the mean, variance, and fitting correlation coefficient are respectively:

[0121]

[0122]

[0123]

[0124] In this example, n is 550 representing the number of sampling points, σ ∈ [0, 2.5] representing the gray value, θ ∈ [0°, 360°] representing the azimuth angle, and r ∈ [0, 4500] (m) representing the radial distance. represents the result of the radar echo intensity after fitting; represents the mean of the radar echo intensity of the selected line; z ∈ [0, 5] represents the radar echo intensity value before fitting.

[0125] Step 1.3: In this example, use the RBF neural network for training as Figure 4 , and initialize the connection weight vector from the hidden layer to the output layer:

[0126] W m = [w m1 , w m2 , … w mq T , (m = 1, 2, …, q)

[0127] In the formula, q is 300, representing the number of hidden layer units. m represents the m-th layer of the hidden layer.

[0128] The method based on center initialization gives the weight initialization formula from the hidden layer to the output layer:

[0129]

[0130] In the formula, mink is the minimum value -1 of the prediction result of the k-th output neuron in the training samples; maxk is the maximum value 1 of the prediction result of the k-th output neuron in the training samples, and j represents the j-th neuron in the hidden layer.

[0131] Step 1.2, initialize the center parameter vector of each neuron in the hidden layer:

[0132] C j = [c j1 , c j2 , … c jn T

[0133] In the formula, j represents the j-th neuron in the hidden layer, and n is 550, representing the number of single scan line sampling points.

[0134] The initial value of the center parameter of the RBF neural network is:

[0135]

[0136] In the formula, mini is the minimum value 0 of all input information of the i-th feature in the training set, and maxi is the maximum value 1 of all input information of the i-th feature in the training set. p is 3, representing the total number of neurons in the hidden layer.

[0137] Step 1.3, initialize the width vector formula:

[0138] D j = [d j1 , d j2 , … d jn T

[0139] In the formula, j represents the j-th neuron in the hidden layer, and n is 550, representing the number of single scan line sampling points.

[0140] Width vector: ​​​

[0141]

[0142] where d f is the width adjustment coefficient of 550, x ij represents the i-th eigenvalue of the j-th scan line, c ij represents the central parameter value. n is the number of sampling points of a single scan line, which is 550.

[0143] Calculation formula for the output of the j-th hidden layer neuron:

[0144]

[0145] where X represents the 3×5559 input feature vector, C represents the 1×550 central parameter vector, and D represents the 1×550 width vector.

[0146] Step 1.4: The input radar image feature data is:

[0147]

[0148] Calculate the radial basis function as the node activation function, and its formula is as follows:

[0149]

[0150] where R∈[0,1] represents the hidden layer neuron value.

[0151] Step 1.5, finally, the formula for calculating the result of the neuron output by the above formula:

[0152]

[0153] where y k is a specific output element in the output vector Y. Among them, the output vector and the expected output vector are respectively:

[0154] Y = [y1, y2,... y q T

[0155] O = [o1, o2,... o q T

[0156] In this example, q is 330, representing the number of hidden layer units.

[0157] Step 1.6, the RBF neural network uses the gradient descent method to adaptively adjust to the optimal value through learning. The iterative calculation formulas for adjusting the weight parameters, centers, and widths are respectively:

[0158]

[0159] ​​

[0160]

[0161] Where \(W\) represents a \(1\times300\) training weight vector, \(C\) represents a \(1\times550\) center parameter vector, and \(D\) represents a \(1\times550\) width vector.

[0162] In step 1.7, after the center, width, and weights are updated, in this example, the values of \(\eta\) and \(\alpha\) are 1 and 3 respectively, and the value of the iteration termination accuracy \(\varepsilon\) is taken as 0.005. The calculation formula for the root mean square error \(RMS\) of the network output is:

[0163]

[0164] Where \(O\) represents the output 0 matrix, and \(N = 5559\) represents the number of scan lines. If \(RMS\leq\varepsilon\), the training ends; otherwise, iterative calculation is performed according to step 1.6, and the weight, center, and width parameters are iteratively calculated until the condition \(RMS\leq\varepsilon\) is satisfied.

[0165] Step 2. Obtain the echo data to be measured on a single scan line in the radar image as attached Figure 5 , in this example, three characteristic parameters, namely the mean, variance, and fitting correlation coefficient of the radar echo data, are selected according to step 1.2, and the characteristic values are calculated. The mean, variance, and fitting correlation coefficient are respectively:

[0166]

[0167]

[0168]

[0169] In this example, \(n = 550\) represents the number of sampling points, \(\sigma\in[0,2.5]\) represents the gray value, \(\theta\in[0^{\circ},360^{\circ}]\) represents the azimuth angle, \(r\in[0,4500](m)\) represents the radial distance, represents the result of the radar echo intensity after fitting; represents the mean of the radar echo intensity of the selected line; \(z\in[0,5]\) represents the radar echo intensity value before fitting.

[0170] The mean value calculated in this example is 0.4779, the variance is 0.6685, and the fitting correlation coefficient is 0.5148.

[0171] Finally, the input characteristic matrix obtained in this example is \([0.4779, 0.6685, 0.5148]\).

[0172] Step 3, input the characteristic matrix calculated in step 2 into the RBF classifier model trained in step 1, and finally obtain 1 output \(y\) value. The output \(y\) value in this example is -0.6.

[0173] Step 4, determine whether there is a target in the selected scan line:

[0174] Compare the finally obtained output y value with the expected output value of 0. If the output is greater than 0, there is a target; if the output is less than 0, there is no target. In this example, the finally obtained output y value of the selected scan line is -0.6, which is less than the expected output value of 0, and it is judged that there is no target.

[0175] During 2016 - 2018, a large amount of radar data and sea condition information for relevant periods were obtained during the navigation of the experimental ship in the East China Sea. Detection was carried out using the method of this article and traditional methods. Statistics were made on the classifier in the three - feature - parameter RBF neural network as Figure 7 . Then, statistics were made on taking the intersection point of the target radar echo probability density curve and the pure sea clutter probability density curve as the threshold by CFAR to divide data samples, and the performance indicators are shown in Table 2.

[0176] Table 2 Performance indicators of the three - feature classifier

[0177]

[0178] As can be seen from Table 2, the three - feature RBF neural network classifier has better detection performance, higher precision and accuracy, and shorter detection time than the CFAR classifier.

[0179] The fast detection algorithm for marine radar targets based on RBF neural network proposed by the present invention is not only truly effective in actual measurement, but also has higher precision and accuracy than the results of CFAR detection technology. It is more consistent with the actual target situation and is more superior to the classical CFAR detection technology in terms of total detection time, which can ensure the safe navigation of ships.

Claims

1. A method for rapid detection of marine radar targets, characterized in that, It includes the following steps: Step 1: Select radar echo data samples on a single scan line from the off-line radar data. The samples include samples containing target radar echoes and samples of pure sea clutter radar echoes without targets; Select characteristic parameters, input the echo eigenvalue of the samples into the RBF neural network for training, and obtain the trained RBF neural network classifier model; Step 1.1: Initialize the connection weight vector from the hidden layer to the output layer: W m = [w m1 , w m2 , … w mq T , m = 1, 2, …, q​ In the formula, q is the number of hidden layer units, and m represents the m-th layer of the hidden layer; The weight initialization formula from the hidden layer to the output layer is: In the formula, mink is the minimum value of the predicted result of the k-th neuron output in the training samples; maxk is the maximum value of the predicted result of the k-th neuron output in the training samples, and j represents the j-th neuron in the hidden layer; Step 1.2: Initialize the central parameter vector of each neuron in the hidden layer: C j = [c j1 , c j2 , … c jn T ​ In the formula, j represents the j-th neuron in the hidden layer, and n represents the number of sampling points on a single scan line; The initial value of the central parameter of the RBF neural network is: In the formula, min i is the minimum value of all input information of the i-th feature in the training set, max i is the maximum value of all input information of the i-th feature in the training set, and p represents the total number of neurons in the hidden layer; Step 1.3: Initialize the width vector formula: D j = [d j1 , d j2 , … d jn T ​ In the formula, j represents the j-th neuron in the hidden layer, and n represents the number of sampling points on a single scan line; Width vector: where d f is the width adjustment coefficient, x ij represents the i-th eigenvalue of the j-th scan line, and c ij represents the central parameter value; The output calculation formula of the j-th hidden layer neuron: In the formula, X represents the input feature vector, C represents its central parameter vector, and D represents the width vector; Step 1.4: The input radar image feature data is: Calculate the radial basis function as the node activation function: In the formula, R is the value of the hidden layer neuron; Step 1.5: Calculate the neuron output: where y k is a specific output element in the output vector Y, where the output vector is: Y = [y1, y2, … y q T ​ The expected output vector is a zero vector, specifically: O = [o1, o2, … o q T ​ In the formula, q represents the number of hidden layer units; Step 1.6: The RBF neural network adaptively adjusts the RBF neural network parameters, namely the connection weight vector, the central parameter vector, and the width vector, to the optimal values by the gradient descent method; the calculation formulas of the connection weight vector, the central parameter vector, and the width vector are respectively: In the formula, W represents the training weight vector, C represents its central parameter vector, and D represents the width vector; Step 1.7: After the center, width, and weight are updated, according to the given values of η and α and the iteration termination accuracy ε value, the calculation formula of the root mean square error RMS value of the network output: In the formula, O represents the output expectation matrix, N is the number of scan lines. If RMS ≤ ε, the training ends; otherwise, return to Step 1.6 to perform iterative calculations on the connection weight vector, the central parameter vector, and the width vector; Step 2: Obtain the echo data to be measured on a single scan line in the radar image, calculate the echo eigenvalue of the echo data to be measured according to the characteristic parameters selected in Step 1, and obtain the input feature matrix; Step 3: Input the input feature matrix obtained in Step 2 into the trained RBF neural network classifier model to obtain the decision function value y; Step 4: Compare the obtained decision function value y with the classifier threshold to determine whether there is a target on the scan line: if y is greater than the threshold, there is a target; if y is less than the threshold, there is no target.

2. A method for rapid detection of marine radar targets according to claim 1, characterized in that: The feature parameters include: mean, variance, and fitting correlation coefficient.

3. A method for rapid detection of marine radar targets according to claim 1, characterized in that: The radar is an X-band marine radar.

Images