A radar target recognition method and system based on feature extraction and matching network

Abstract

This invention discloses a radar target recognition method based on feature extraction and matching networks, comprising the following steps: a) decomposing the radar echo EMD to obtain its first intrinsic mode function imf1 and the sum of the remaining intrinsic mode functions imf r b. The fuselage component x is obtained by using the CLEAN algorithm to sum all intrinsic moduli functions obtained after EMD decomposition. F and micro-motion component x JEM c. Add all components imf1 and imf r x F and x JEM All images are converted using Gram's angle field transform; d) all images are stitched together, and a matching network is used to identify the stitched images to obtain classification results. This invention integrates domain expert knowledge to extract some features and then uses a deep neural network for classification, which can reduce the learning difficulty and improve the prediction accuracy of the neural network.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic information technology, and in particular to a radar target recognition method and system based on feature extraction and matching network. BACKGROUND

[0002] In modern military operations, having air superiority often represents having absolute initiative in the battlefield. Helicopters are agile in action, suitable for complex terrain, and other advantages make them irreplaceable equipment in air battlefield. Propeller aircraft has superior low-altitude and ultra-low-altitude maneuverability, mainly implements low-speed multipurpose combat of battlefield control and battlefield support, and has considerable integrated combat capability. Jet aircraft has high aircraft speed and maneuverability, mainly undertakes tasks such as high-altitude capture of air superiority in the war zone and precise attack on enemy fixed targets, and is an important core force of air combat. The three types of aircraft bear different military tasks in the air battlefield, and make great contributions to winning the war. Therefore, it is of great significance to carry out classification and recognition of the three types of aircraft, i.e. helicopters, propellers and jets. The radar target recognition method based on traditional classifier generally includes two tasks: firstly, the characteristic information of target echo is used to extract target features through feature engineering, and then a suitable classifier is selected to classify the extracted feature information.

[0003] The above method has the following defects: the effectiveness of target features and the performance of the classifier are important factors affecting the final recognition result, and the recognition effect is greatly affected by the environment, and the generalization also has defects, which makes it difficult to meet the requirements of self-adaptation in actual recognition scene. SUMMARY

[0004] The present application aims to provide a radar target recognition method based on feature extraction and matching network, which uses feature engineering to extract key feature frequencies in radar echo, and innovatively uses Gram angle field to convert one-dimensional echo data into images. The Empirical Mode Decomposition (EMD) algorithm is used for narrowband radar echo data to separate the fuselage component and the micro-motion component of jet aircraft and helicopter. The CLEAN algorithm is used to remove the maximum harmonic in the signal, thereby separating the fuselage component and the micro-motion component of propeller aircraft. Then all the separated signals are converted into images through the Gram angle field, and all the images are spliced and put into the matching network for reasoning to obtain the type.

[0005] Deep learning, with its excellent generalization and adaptive capabilities, effectively addresses the aforementioned problems of traditional recognition methods and is increasingly being widely applied in radar target recognition. Leveraging the powerful adaptive learning capabilities of deep learning, feature extraction models applicable to a wide range of radar targets and scenarios can be found. Furthermore, deep networks combine feature extraction with target recognition to obtain an integrated learning and classification model. Therefore, it is necessary to explore new methods to solve the problem of target recognition using limited radar echo data.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A radar target recognition method based on feature extraction and matching networks includes the following steps:

[0008] a. The radar echo EMD decomposition yields its first eigenmode function imf1 and the sum of the remaining eigenmode functions imf. r ;

[0009] b. The fuselage component x is obtained by applying the CLEAN algorithm to the sum of all intrinsic mode functions obtained after EMD decomposition. F and micro-motion component x JEM ;

[0010] c. Add all components imf1 and imf r x F and x JEM All were transformed into images using the Gram angular field method;

[0011] d. Stitch all the images together, and use a matching network to identify the stitched images and obtain the classification results;

[0012] Preferably, in step a, the EMD decomposition of the radar echo x(t) is expressed as: imf i Represents the i-th intrinsic mode signal, r L The remaining terms after L decompositions are as follows:

[0013] a.1 Initialization: Let r(t) = x(t);

[0014] a.2 Extract all local maxima of r(t), interpolate the sequence of maxima to obtain the upper envelope r of r(t). max (t);

[0015] a.3 Extract all local minima of r(t), interpolate the minima sequence to obtain the lower envelope r(t). min (t);

[0016] a.4, for r at each time stepmax (t) and r min Calculate its average value (t) to obtain the instantaneous average value m(t): m(t) = (r max (t)+r min (t)) / 2;

[0017] a.5. Let r(t) = r(t) - m(t);

[0018] a.6. Determine whether r(t) satisfies the two conditions of IMF. If so, the first IMF is obtained and denoted as imf1. Otherwise, repeat the operation from step a.2 to step a.5 on the r(t) obtained in step a.5 until r(t) satisfies the two conditions of IMF and imf1 is obtained.

[0019] Preferably, in step b, the remainder r obtained from EMD decomposition L This corresponds to the near-ground clutter component, so the signal after clutter suppression can be obtained as follows: This signal can be represented as harmonic components in the following form:

[0020] The maximum harmonic component x corresponding to the fuselage component is obtained using the CLEAN algorithm. F and micro-motion component x JEM ,specific

[0021] The steps are as follows:

[0022] b.1 Input time-domain signal x EMD ;

[0023] b.2 Calculate the Doppler spectrum of the time-domain signal X = FFT(x EMD )

[0024] b.3 Find the maximum value point in X and calculate its Doppler frequency f1 based on the position of the maximum value point. Record the corresponding amplitude ρ1 and phase.

[0025] b.4. Based on the results of b.3, reconstruct the maximum harmonic component in the signal to obtain the time corresponding to the fuselage component.

[0026] Domain echo; b.5. From the original signal x EMD Subtracting the maximum harmonic component reconstructed in step b.4, we obtain the time-domain echo x corresponding to the micro-motion component. JEM =x EMD -x F .

[0027] Preferably, in step c, the input signal is first scaled to the range [-1, 1] using the following formula: Where xmin and x max These are the maximum and minimum values ​​of the signal. It is the scaled x(t);

[0028] Then, the rectangular coordinate system is converted to the polar coordinate system using the following formula.

[0029]

[0030] By calculating the sum of the cosine functions at each time point, the two-dimensional matrix obtained through the Gram angle field transformation is shown in the following equation.

[0031]

[0032] Preferably, the matching network incorporates an attention mechanism to calculate the predicted output as follows:

[0033]

[0034] Where x i y i Indicates support set Samples and labels, Represents query set D q The sample, This represents the predicted output of the matching network. Its attention mechanism can be represented by the following equation:

[0035]

[0036] Where cosine represents the cosine distance, f(·) and g(·) are the embedding functions implemented by the convolutional neural network, and x i The sample represents the support set. Represents a sample of the query set;

[0037] The objective function of the matching network is shown below:

[0038]

[0039] Where θ represents the parameters of the matching network, and D represents the meta-training set. s and D q These represent the support set and query set, respectively, of the meta-training set.

[0040] This invention also discloses a radar target recognition system based on feature extraction and matching networks, comprising: an EMD decomposition module for performing EMD decomposition on radar echoes, to its first intrinsic mode function imf1 and the sum of the remaining intrinsic mode functions imf r ;

[0041] The CLEAN algorithm module is used to perform the CLEAN algorithm on the sum of all intrinsic mode functions obtained after EMD decomposition, thereby obtaining the fuselage component x. F and micro-motion component x JEM ;

[0042] The Gram corner field module is used to store all components imf1, imf r x F and x JEM Transform into an image;

[0043] The image stitching module is used to stitch together all the components imf1, imf... r x F and x JEM The transformed images are stitched together;

[0044] The matching network module is used to identify and classify the stitched images.

[0045] The beneficial effects of this invention are as follows:

[0046] 1. This invention integrates the knowledge of experts in the field to extract some features and then uses a deep neural network for classification, which can reduce the learning difficulty and improve the prediction accuracy of the neural network.

[0047] 2. This invention uses a matching network instead of the traditional LSTM or CNN as a classifier, which improves the learning ability for small samples and can achieve good results even when there is little military data. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of a radar target recognition method based on feature extraction and matching networks.

[0049] Figure 2 This is a structural diagram of a radar target recognition system based on feature extraction and matching networks. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings.

[0051] Example 1

[0052] This embodiment discloses a radar target recognition method, such as Figure 1 As shown, this radar target recognition method is a radar target recognition method based on feature extraction and matching networks, as detailed below:

[0053] 1. Adopting such Figure 1The system architecture shown is a radar target recognition system based on feature extraction and matching networks. First, the radar echo EMD is decomposed to obtain its first intrinsic mode function (imf1) and the sum of the remaining intrinsic mode functions (imf). r The fuselage component x can be obtained by applying CLEAN to the sum of all intrinsic moduli obtained after EMD decomposition. F and micro-motion component x JEM All components imf1, imf r x F and x JEM All images were generated using Gram angular field transform, then stitched together, and finally a matching network was used to identify and classify the stitched images.

[0054] 2. Feature extraction based on EMD-CLEAN algorithm

[0055] 1) Feature Analysis: Jet aircraft fly much faster than helicopters. In their radar echoes, the fuselage component corresponds to the high-frequency component, while the slower flight speed and rapidly rotating propeller of helicopters result in the fuselage component corresponding to the low-frequency component. The fuselage component of propeller-driven aircraft is much larger than the micro-motion component.

[0056] The maximum harmonic component can be used to correspond to the fuselage component.

[0057] 2) EMD decomposition (extracting fuselage and micro-motion components from jet and helicopter aircraft): The EMD decomposition of the original signal x(t) can be expressed as... imf i Represents the i-th intrinsic mode signal, r L This represents the remaining terms after L decompositions. The decomposition process is as follows:

[0058] ① Initialization: Let r(t) = x(t);

[0059] ② Extract all local maxima of r(t), interpolate the sequence of maxima to obtain the upper envelope r of r(t). max (t);

[0060] ③ Extract all local minima of r(t), interpolate the minima sequence to obtain the lower envelope r of r(t). min (t);

[0061] ④ For r at each time step max (t) and r min Calculate its average value (t) to obtain the instantaneous average value m(t):

[0062] m(t)=(r max (t)+r min(t)) / 2;

[0063] ⑤ Let r(t) = r(t) - m(t);

[0064] ⑥ Determine whether r(t) satisfies the two conditions of IMF. If so, the first IMF is obtained and denoted as imf1. Otherwise, repeat the operation from step ② to step ⑤ on r(t) obtained in step ⑤ until r(t) satisfies the two conditions of IMF and imf1 is obtained.

[0065] 3) CLEAN algorithm for extracting maximum harmonics: the remainder r obtained from EMD decomposition L This corresponds to the near-ground clutter component, so the signal after clutter suppression can be obtained as follows: This signal can be represented as harmonic components and in the form of... The CLEAN algorithm can obtain the maximum harmonic component x corresponding to the fuselage component. F and micro-motion component x JEM The specific steps are as follows:

[0066] ① Input time-domain signal x EMD ;

[0067] ② Calculate the Doppler spectrum of the time-domain signal X = FFT(x) EMD );

[0068] ③ Find the maximum value point from X and calculate its Doppler frequency f1 based on the position of the maximum value point. Record the corresponding amplitude ρ1 and phase.

[0069] ④ Based on the results of step ③, reconstruct the maximum harmonic component in the signal to obtain the time-domain echo corresponding to the fuselage component;

[0070] ⑤ From the original signal x EMD Subtracting the maximum harmonic component reconstructed in step ④ yields the time-domain echo x corresponding to the micro-motion component. JEM =x EMD -x F .

[0071] 3. Calculation of Gram angle field: First, scale the input signal to the range [-1, 1] using the following formula:

[0072]

[0073] Where x min and x max These are the maximum and minimum values ​​of the signal. This is the scaled x(t). Then, the Cartesian coordinate system is converted to polar coordinates using the following formula.

[0074]

[0075] By calculating the sum of the cosine functions at each time point, the two-dimensional matrix obtained through the Gram angle field transformation can be shown in the following equation.

[0076]

[0077] 4. Matching Network Processing: The matching network introduces an attention mechanism to calculate the predicted output as follows:

[0078]

[0079] Where x i y i Indicates support set Samples and labels, Represents query set D q The sample, This represents the predicted output of the matching network. Its attention mechanism can be expressed by the following equation:

[0080]

[0081] Where cosine represents the cosine distance, f(·) and g(·) are the embedding functions implemented by the convolutional neural network, and x i The sample represents the support set. This represents a sample in the query set. The objective function of the matching network is shown below:

[0082]

[0083] Where θ represents the parameters of the matching network, and D represents the meta-training set. s and D q These represent the support set and query set, respectively, of the meta-training set.

[0084] Example 2

[0085] Based on Example 1, this embodiment discloses a radar target recognition system based on feature extraction and matching networks. This radar target recognition system based on feature extraction and matching networks adopts the radar target recognition method based on feature extraction and matching networks as described in Example 1. Figure 2 As shown, the system includes:

[0086] The EMD decomposition module is used to perform EMD decomposition on radar echoes, decomposing them into their first eigenmode function imf1 and the sum of the remaining eigenmode functions imf. r ;

[0087] The CLEAN algorithm module is used to perform the CLEAN algorithm on the sum of all intrinsic mode functions obtained after EMD decomposition, thereby obtaining the fuselage component x.F and micro-motion component x JEM ;

[0088] The Gram corner field module is used to store all components imf1, imf r x F and x JEM Transform into an image;

[0089] The image stitching module is used to stitch together all the components imf1, imf... r x F and x JEM The transformed images are stitched together;

[0090] The matching network module is used to identify and classify the stitched images.

[0091] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A radar target recognition method based on feature extraction and matching networks, characterized in that, Includes the following steps: a. The radar echo EMD decomposition yields its first eigenmode function imf1 and the sum of the remaining eigenmode functions imf. r ; b. The fuselage component x is obtained by applying the CLEAN algorithm to the sum of all intrinsic mode functions obtained after EMD decomposition. F and micro-motion component x JEM ; c. Add all components imf1 and imf r x F and x JEM All were transformed into images using the Gram angular field method; d. Stitch all the images together, and use a matching network to identify the stitched images and obtain the classification results.

2. The radar target recognition method based on feature extraction and matching networks according to claim 1, characterized in that, In step a, the EMD decomposition of the radar echo x(t) is expressed as: imf i Represents the i-th intrinsic mode signal, r L The remaining terms after L decompositions are as follows: a.1 Initialization: Let r(t) = x(t); a.2 Extract all local maxima of r(t), interpolate the sequence of maxima to obtain the upper envelope r of r(t). max (t); a.3 Extract all local minima of r(t), interpolate the minima sequence to obtain the lower envelope r(t). min (t); a.4, for r at each time step max (t) and r min Calculate its average value (t) to obtain the instantaneous average value m(t): m(t) = (r max (t)+r min (t)) / 2; a.

5. Let r(t) = r(t) - m(t); a.

6. Determine whether r(t) satisfies the two conditions of IMF. If so, the first IMF is obtained and denoted as imf1. Otherwise, repeat the operation from step a.2 to step a.5 on the r(t) obtained in step a.5 until r(t) satisfies the two conditions of IMF and imf1 is obtained.

3. The radar target recognition method based on feature extraction and matching networks according to claim 2, Its features are, In step b, the remainder r obtained from EMD decomposition L This corresponds to the near-ground clutter component, so the signal after clutter suppression can be obtained as follows: This signal can be represented as harmonic components in the following form: The maximum harmonic component x corresponding to the fuselage component is obtained using the CLEAN algorithm. F and micro-motion component x JEM The specific steps are as follows: b.1 Input time-domain signal x EMD ; b.2 Calculate the Doppler spectrum of the time-domain signal X = FFT(x EMD ) b.3 Find the maximum value point in X and calculate its Doppler frequency f1 based on the position corresponding to the maximum value point. Record the corresponding amplitude ρ1 and phase b.

4. Based on the results of b.3, reconstruct the maximum harmonic component in the signal to obtain the time-domain echo corresponding to the fuselage component; b.

5. From the original signal x EMD Subtracting the maximum harmonic component reconstructed in step b.4, we obtain the time-domain echo x corresponding to the micro-motion component. JEM =x EMD -x F .

4. The radar target recognition method based on feature extraction and matching networks according to claim 3, characterized in that, In step c, the input signal is first scaled to the range [-1, 1] using the following formula: Where x min and x max These are the maximum and minimum values ​​of the signal. It is the scaled x(t); Then, the rectangular coordinate system is converted to the polar coordinate system using the following formula. By calculating the sum of the cosine functions at each time point, the two-dimensional matrix obtained through the Gram angle field transformation is shown in the following equation.

5. The radar target recognition method based on feature extraction and matching networks according to claim 3, characterized in that, The matching network introduces an attention mechanism to calculate the predicted output as follows: Where x i y i Indicates support set Samples and labels, Represents query set D q The sample, This represents the predicted output of the matching network; Its attention mechanism can be represented by the following formula: Where cosine represents the cosine distance, f(·) and g(·) are the embedding functions implemented by the convolutional neural network, and x i The sample represents the support set. Represents a sample of the query set; The objective function of the matching network is shown below: Where θ represents the parameters of the matching network, and D represents the meta-training set. s and D q These represent the support set and query set, respectively, of the meta-training set.

6. A radar target recognition system based on feature extraction and matching networks, characterized in that, include: The EMD decomposition module is used to perform EMD decomposition on radar echoes, obtaining its first eigenmode function imf1 and the sum of the remaining eigenmode functions imf. r ; The CLEAN algorithm module is used to perform the CLEAN algorithm on the sum of all intrinsic mode functions obtained after EMD decomposition, thereby obtaining the fuselage component x. F and micro-motion component x JEM ; The Gram corner field module is used to store all components imf1, imf r x F and x JEM Transform into an image; The image stitching module is used to stitch together all the components imf1, imf... r x F and x JEM The transformed images are stitched together; The matching network module is used to identify and classify the stitched images.

Images