Target detection method and device based on radar echo time sequence dependency relationship

Abstract

The invention provides a target detection method and device based on a radar echo time sequence dependency relationship, and the method comprises the steps: carrying out the preprocessing of a radar echo sequence set, and obtaining a preprocessed radar echo sub-sequence set; carrying out real-number processing and smoothing processing on each sub-sequence in the preprocessed radar echo sub-sequence set one by one to obtain a smoothed radar echo sub-sequence set; obtaining a corresponding optimization coefficient of the feature representation learner based on the first-order partial order relationship; performing feature extraction on corresponding subsequences in the smoothed radar echo subsequence set one by one to obtain a feature set keeping a first-order partial order relation; training a detector based on the feature set keeping the first-order partial order relationship and the target identifier to obtain a trained detector; and detecting the radar echo sequence to be detected to obtain a target prediction result. The technical problem that a radar target detection method in the prior art lacks echo feature representation learning or is unbalanced in detection can be solved.

Description

Technical Field

[0001] The present invention relates to the field of radar target detection technology, and in particular to a target detection method and device based on radar echo time sequence dependency. Background Art

[0002] As a detection device widely used in the fields of social science development and military defense, radar has the characteristics of good robustness in extreme weather and insensitivity to light, and has important academic and practical value. The detection performance of the classic radar target detection method depends on a lot of manual parameter adjustment. Benefiting from the development of big data science, machine learning methods have been gradually applied to the field of radar target detection and have shown many advantages in terms of robustness and generalization.

[0003] The mainstream radar target detection algorithms based on machine learning can be divided into target-clutter detection methods based on echo signals, target detection-recognition methods based on range / Doppler / angle frequency domain transformation, and target recognition methods based on radar imaging data. However, existing research lacks the discussion of learning methods for echo feature representation, and faces the problem of unbalanced detection caused by extremely sparse available target information. Summary of the invention

[0004] In order to solve one of the above technical problems, the present invention provides a target detection method and device based on the temporal dependency of radar echoes, which can solve the technical problems in the prior art that radar target detection methods lack learning of echo feature representation or detection imbalance.

[0005] According to one aspect of the present invention, a target detection method based on radar echo temporal dependency is provided, the method comprising:

[0006] Preprocessing the radar echo sequence set to obtain a preprocessed radar echo subsequence set;

[0007] Performing real number conversion and smoothing processing on each subsequence in the preprocessed radar echo subsequence set one by one to obtain a smoothed radar echo subsequence set;

[0008] For each subsequence in the smoothed radar echo subsequence set, the corresponding optimization coefficient of the feature representation learner is obtained based on the first-order partial order relationship;

[0009] A feature representation learner with an optimized coefficient is used to extract features from the corresponding subsequences in the smoothed radar echo subsequence set one by one, and a feature set that maintains a first-order partial order relationship is obtained;

[0010] The detector is trained based on the feature set and target identification that maintain the first-order partial order relationship to obtain a trained detector;

[0011] Based on the trained detector, the radar echo sequence to be tested is detected to obtain the target prediction result.

[0012] Preferably, for each subsequence in the smoothed radar echo subsequence set, obtaining a corresponding optimization coefficient of a feature representation learner based on a first-order partial order relationship includes:

[0013] The expression for obtaining the feature representation learner based on the first-order partial order relationship is as follows:

[0014]

[0015] Obtaining a corresponding optimization coefficient of the feature representation learner based on the expression of the feature representation learner;

[0016] In the formula, is the feature representation learner, is a subsequence in the smoothed radar echo subsequence set, are the first, tth, and Tth elements in the current subsequence respectively, and T is the number of distance units. The energy function for reconstructing the first-order partial order relation for the measurement model, for The metric function of , e is a learnable parameter, is a first-order partial order relation, are the i-th and j-th elements in the current subsequence respectively, They are The measurement function of .

[0017] Preferably, obtaining a corresponding optimization coefficient of the feature representation learner based on an expression of the feature representation learner includes:

[0018] If the metric function is defined as a linear function, the constraint condition of the expression of the feature representation learner is as follows:

[0019] The constraints are regressed element by element After decoupling, the expression of the transformed feature representation learner is as follows:

[0020]

[0021] The expression of the transformed feature representation learner is relaxed and regularized, and the expression of the processed feature representation learner is as follows:

[0022]

[0023] The processed feature represents a subsequence in the smoothed radar echo subsequence set in the learner Nonlinearly mapping from low-dimensional space to high-dimensional space, the expression of the mapped feature representation learner is as follows:

[0024]

[0025] Solve the expression of the mapped feature representation learner to obtain the optimal solution of the feature representation learner, and use the optimal solution of the feature representation learner as the subsequence of the feature representation learner. The corresponding optimization coefficient;

[0026] In the formula, is the transformed feature representation learner, ||·|| is a norm operator, is the processed feature representation learner, is the L2 regularization term, ε is the relaxation factor of the regression error term, λ is the penalty factor of the regression error term, [·] ≥0 =max{·;0} is the error loss term that only cares about the regression results that exceed the error relaxation boundary, Φ(V) is the feature representation learner after mapping, is a nonlinear mapping function.

[0027] Preferably, after preprocessing the radar echo sequence set to obtain the preprocessed radar echo subsequence set, the method further includes: upsampling the target sample set in the preprocessed radar echo subsequence set and downsampling the clutter sample set to obtain the radar echo subsequence set after equalization processing.

[0028] Preferably, upsampling the target sample set in the preprocessed radar echo subsequence set comprises:

[0029] S1, vectorizing the target sample set in the preprocessed radar echo subsequence set to obtain a vectorized target sample set;

[0030] S2, randomly select a target sample from the vectorized target sample set with replacement, and calculate a nearest neighbor samples of the randomly selected target sample;

[0031] S3, randomly select a neighbor sample from a neighbor samples for vectorization, and obtain the vectorized neighbor sample;

[0032] S4, performing vector slicing restoration on the vectorized neighbor samples to obtain restored neighbor samples;

[0033] S5, taking the restored neighbor samples as new target samples and putting them into the generated target sample set;

[0034] S6. Repeat S2-S5 L×M times to obtain a generated target sample set having L×M new target samples, and put the generated target sample set into the target sample set in the preprocessed radar echo subsequence set to obtain the target sample set after equalization processing, thereby completing the upsampling of the target sample set, wherein L is the synthesis multiple, and M is the number of target echo subsequences in the preprocessed radar echo subsequence set.

[0035] Preferably, preprocessing the radar echo sequence set to obtain a preprocessed radar echo subsequence set includes:

[0036] Remove abnormal points from the radar echo sequence set to obtain a radar echo sequence set without abnormal points;

[0037] Performing digital pulse compression on a radar echo sequence set without abnormal points to obtain a compressed radar echo sequence set;

[0038] Performing finite impulse response filtering on the compressed radar echo sequence set to obtain a filtered radar echo sequence set;

[0039] The filtered radar echo sequence set is segmented by a sliding window to obtain a preprocessed radar echo subsequence set.

[0040] Preferably, performing digital pulse compression on the radar echo sequence set without abnormal points to obtain the compressed radar echo sequence set includes:

[0041] Perform Fourier transform on the radar echo sequence set without abnormal points channel by channel to obtain the frequency domain response curve of the pulse corresponding to each channel.

[0042] Performing complex restoration on the original reference transmission signal to obtain a restored reference transmission signal;

[0043] The frequency domain response curve of each pulse is multiplied bit by bit with the restored reference transmission signal and inverse Fourier transformed channel by channel to obtain a compressed radar echo sequence set;

[0044] The compressed radar echo sequence set is subjected to finite impulse response filtering, and the filtered radar echo sequence set includes:

[0045] The compressed radar echo sequence set is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence set:

[0046]

[0047] In the formula, Z p (n) is the filtered radar echo sequence of the nth range unit of the pth channel, K is the total number of filters, Hk,p is the coefficient matrix of the p-th channel of the k-th filter, is the transpose of the compressed radar echo sequence of the nth range unit of the kth filter.

[0048] Preferably, detecting the radar echo sequence to be detected based on the trained detector to obtain the target prediction result includes:

[0049] Preprocessing the radar echo sequence to be measured to obtain a preprocessed radar echo subsequence to be measured;

[0050] The preprocessed radar echo subsequence to be measured is converted into real numbers and smoothed to obtain a smoothed radar echo subsequence to be measured;

[0051] The feature representation learner with temporal feature representation is used to extract the features of the smoothed radar echo subsequence to be measured, and the features to be measured that maintain the first-order partial order relationship are obtained.

[0052] The features to be tested that maintain the first-order partial order relationship are input into the trained detector to detect the radar echo sequence to be tested, and the target prediction result is obtained.

[0053] Preferably, preprocessing the radar echo sequence to be measured to obtain the preprocessed radar echo subsequence to be measured includes:

[0054] Remove abnormal points from the radar echo sequence to be tested, and obtain a radar echo sequence to be tested without abnormal points;

[0055] Performing digital pulse compression on the radar echo sequence to be tested without abnormal points to obtain a compressed radar echo sequence to be tested;

[0056] Performing finite impulse response filtering on the compressed radar echo sequence to be measured to obtain a filtered radar echo sequence to be measured;

[0057] The filtered radar echo sequence to be measured is segmented by a sliding window to obtain a preprocessed radar echo subsequence to be measured.

[0058] Preferably, performing digital pulse compression on the radar echo sequence to be measured without abnormal points to obtain the compressed radar echo sequence to be measured includes:

[0059] Perform Fourier transform on the radar echo sequence to be tested without abnormal points, and obtain the frequency domain response curve of the pulse corresponding to each channel.

[0060] Performing complex restoration on the original reference transmission signal to obtain a restored reference transmission signal;

[0061] The frequency domain response curve of each pulse is multiplied bit by bit with the restored reference transmission signal and inverse Fourier transformed channel by channel to obtain a compressed radar echo sequence to be tested;

[0062] The compressed radar echo sequence to be measured is subjected to finite impulse response filtering, and the filtered radar echo sequence to be measured includes:

[0063] The compressed radar echo sequence to be measured is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence to be measured:

[0064]

[0065] In the formula, Z p ′(n) is the radar echo sequence to be measured after filtering of the nth range unit of the pth channel, K is the total number of filters, H k,p is the coefficient matrix of the p-th channel of the k-th filter, is the transpose of the compressed radar echo sequence to be measured of the nth range unit of the kth filter.

[0066] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a target detection program based on radar echo timing dependency stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned methods when executing the target detection program based on radar echo timing dependency.

[0067] By applying the technical solution of the present invention, the radar echo sequence set is preprocessed to obtain an echo signal with a higher signal-to-noise ratio; the temporal feature representation of the feature representation learner is obtained through the first-order partial order relationship, thereby obtaining a feature representation learner with temporal feature representation, and the feature representation learner is used to extract features and perform subsequent training, thereby realizing target detection using the temporal dependency of radar echoes, and solving the problem that the radar target detection method in the prior art lacks learning of echo feature representation. The present invention can be used in different task fields such as perception, control, and decision-making, and can also be directly applied to radar equipment with various detection systems such as continuous wave and pulse Doppler. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0069] Figure 1A flow chart of a target detection method based on radar echo timing dependency provided according to an embodiment of the present invention is shown;

[0070] Figure 2 A block diagram of a target detection method based on radar echo timing dependency provided according to an embodiment of the present invention is shown;

[0071] Figure 3 A schematic diagram of learning temporal feature representation of a feature representation learner provided according to an embodiment of the present invention is shown;

[0072] Figure 4 A schematic diagram of processing target-clutter sample equalization provided according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0073] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0075] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0076] like Figure 1 and Figure 2 As shown, the present invention provides a target detection method based on radar echo time sequence dependency, the method comprising:

[0077] S10, preprocessing the radar echo sequence set to obtain a preprocessed radar echo subsequence set;

[0078] S20, performing real number conversion and smoothing processing on each subsequence in the preprocessed radar echo subsequence set one by one to obtain a smoothed radar echo subsequence set;

[0079] S30, for each subsequence in the smoothed radar echo subsequence set, obtaining a corresponding optimization coefficient of a feature representation learner based on a first-order partial order relationship;

[0080] S40, using a feature representation learner with an optimized coefficient to extract features from corresponding subsequences in the smoothed radar echo subsequence set one by one, to obtain a feature set that maintains a first-order partial order relationship;

[0081] S50, training the detector based on the feature set and the target identification that maintain the first-order partial order relationship to obtain a trained detector;

[0082] S60: Detect the radar echo sequence to be detected based on the trained detector to obtain a target prediction result.

[0083] The present invention preprocesses the radar echo sequence set to obtain an echo signal with a higher signal-to-noise ratio; obtains the temporal feature representation of the feature representation learner through the first-order partial order relationship, thereby obtaining a feature representation learner with temporal feature representation, and uses the feature representation learner to extract features and perform subsequent training, thereby realizing target detection using the temporal dependency of radar echoes, and solving the problem that the radar target detection method in the prior art lacks learning of echo feature representation. The present invention can be used in different task fields such as perception, control, and decision-making, and can also be directly applied to radar equipment with various detection systems such as continuous wave and pulse Doppler.

[0084] According to an embodiment of the present invention, in S10 of the present invention, in order to obtain an echo signal with a higher signal-to-noise ratio and comprehensively consider the echo timing context content and the timeliness of processing, the radar echo sequence set is preprocessed to obtain the preprocessed radar echo subsequence set including:

[0085] S11, removing abnormal points from the radar echo sequence set to obtain a radar echo sequence set without abnormal points;

[0086] S12, performing digital pulse compression on the radar echo sequence set without abnormal points to obtain a compressed radar echo sequence set;

[0087] S13, performing finite impulse response filtering on the compressed radar echo sequence set to obtain a filtered radar echo sequence set;

[0088] S14, performing sliding window segmentation on the filtered radar echo sequence set to obtain a preprocessed radar echo subsequence set.

[0089] Specifically, in S12 of the present invention, digital pulse compression is performed on a radar echo sequence set without abnormal points, and the compressed radar echo sequence set obtained includes:

[0090] S121, performing a channel-by-channel Fourier transform on a radar echo sequence set without abnormal points to obtain a frequency domain response curve of a pulse corresponding to each channel;

[0091] S122, performing complex restoration on the original reference transmit signal to obtain a restored reference transmit signal;

[0092] S123, multiplying the frequency domain response curve of each pulse by the restored reference transmission signal bit by bit and performing inverse Fourier transform on a channel by channel basis to obtain a compressed radar echo sequence set.

[0093] For example, given a radar echo sequence set without abnormal points in channel P The original reference transmission signal Among them, T is the number of distance units, that is, the fast time dimension, and P is the number of pulses, that is, the number of Doppler channels, that is, the slow time dimension. is a complex field, is the real number domain, and N is the number of channels of the reference transmission signal. First, perform a channel-by-channel Fourier transform (FFT) on X to obtain the frequency domain response curve of each pulse Then perform complex reduction on S to obtain the reduced reference transmission signal Then With S * Perform bit-by-bit multiplication; finally, perform channel-by-channel inverse Fourier transform (IFFT) on the multiplied result to obtain a compressed radar echo sequence set.

[0094] In S13 of the present invention, finite impulse response filtering is performed on the compressed radar echo sequence set, and the filtered radar echo sequence set obtained includes:

[0095] The compressed radar echo sequence set is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence set:

[0096]

[0097] In the formula, Z p (n) is the filtered radar echo sequence of the nth range unit of the pth channel, K is the total number of filters, H k,p is the coefficient matrix of the p-th channel of the k-th filter, is the transpose of the compressed radar echo sequence of the nth range unit of the kth filter.

[0098] For example, given a designed P group FIR filter coefficient matrix The transpose of the compressed radar echo sequence set Then the filtered radar echo sequence set can be obtained through the above formula.

[0099] In S14 of the present invention, a sliding window mechanism is used to segment the filtered radar echo sequence set to obtain multiple more compact echo subsequences, which will be used as input for the subsequent time series modeling method. The size of the sliding window depends on the number of reference units used. If the number of reference units is L, the sliding window size is 2L+1. In addition, the step size of the sliding window movement is also a parameter that can be set. In the sequential verification link, since the complete sequence needs to be traversed, the step size is 1.

[0100] According to an embodiment of the present invention, in S30 of the present invention, for each subsequence in the smoothed radar echo subsequence set, obtaining a corresponding optimization coefficient of a feature representation learner based on a first-order partial order relationship includes:

[0101] S31. The expression of the feature representation learner based on the first-order partial order relationship is as follows:

[0102]

[0103] S32, obtaining a corresponding optimization coefficient of the feature representation learner based on the expression of the feature representation learner;

[0104] In the formula, is the feature representation learner, is a subsequence in the smoothed radar echo subsequence set, are the first, tth, and Tth elements in the current subsequence respectively, and T is the number of distance units. It is the energy function for measuring the reconstruction of the first-order partial order relationship by the model. The smaller its value is, the smaller the energy cost of the model to reconstruct the temporal relationship is. for The metric function is intuitively the sequence Each element in is projected from the original data space to the The metric interval determined by is, e is a learnable parameter, is a first-order partial order relation, are the i-th and j-th elements in the current subsequence respectively, They are The measurement function of .

[0105] The present invention uses the temporal properties of radar echoes themselves as constraints. Compared with detection methods that ignore representation learning, the temporal representation learning method can use a more compact echo representation to enable the detection model to achieve better comprehensive detection performance.

[0106] In the present invention, the echo is regarded as a time series signal, in which the time sequence information between each unit has important utilization value. Based on the idea of ​​sequence regression learning with the sequence time sequence as a constraint, the present invention proposes an unsupervised time series feature learning method suitable for radar echoes, so as to obtain a feature representation learner (EchoDarwin) with time series feature representation. Among them, the time series information in the sequence essentially reflects the time sequence dependency of each element of the sequence, and this relationship determines the temporal change trend of the element. Therefore, in order to capture the time series information in the sequence, the optimization of the sequence model needs to satisfy the time dependency, that is, a first-order partial order relationship. This first-order partial order relationship is abstractly represented as Then the expression of the feature representation learner (EchoDarwin) with temporal feature representation can be expressed in the form of step S31.

[0107] Specifically, in S32 of the present invention, as Figure 3 As shown, the corresponding optimization coefficients of the feature representation learner are obtained based on the expression of the feature representation learner, including:

[0108] S321. Quantify the front and back in time by using the small and large values ​​that also have a first-order transfer relationship, and consider a simplest model, that is, define the metric function as a linear function. Then the constraint condition of the expression of the feature representation learner is as follows:

[0109] S322. Ideally, for the sequence If the optimal learnable parameter e that satisfies the above constraints can be found, the complete temporal relationship of the sequence can be captured. The learnable parameter e can be regarded as the normal vector of a hyperplane in the original data space of the sequence. The distance from the plane measures the order of the plane in the time dimension. Therefore, the learnable parameter e can be used as a representation of this order, that is, a time-series feature representation. However, it is difficult to directly use the above constraints of coupled outputs at different times to solve e, so it is necessary to introduce some kind of "agent" mechanism to decouple this pair of constraints with transitive properties. Therefore, the constraints are regressed element by element. Decouple, that is, return f(·; e) to the moment position information of the sequence itself: It can be seen that since the time position naturally satisfies the first-order partial order relationship with transitive properties, the time position regression not only decouples the above pairwise constraints, but also can implicitly embed them into the optimization objective, so that the expression of the transformed feature representation learner is as follows:

[0110]

[0111] S323. In order to deal with the jitter of the echo sequence itself, it is also necessary to relax the strict regression target of the expression of the transformed feature representation learner, and at the same time, regularize the learnable parameter e to improve the generalization of the regression model. To this end, the regression relaxation amount e and the regularization constraint term of e are further introduced, that is, the expression of the transformed feature representation learner is relaxed and regularized, and the expression of the processed feature representation learner is obtained as follows:

[0112]

[0113] The above formula just satisfies the Support Vector Regression (SVR) problem with L2 regularization constraints. In the specific solution process, the Trust Region Newton (TRON) method can be used, which is a general optimization method for solving unconstrained and boundary-constrained differentiable problems.

[0114] S324. To further enhance the temporal coding capability of EchoDarwin, the present invention uses a formally determined nonlinear feature mapping (NFM) operation based on the kernel function idea to map the input features in the original space to a higher-dimensional target space. This approach is equivalent to calculating the nonlinear kernel function in the original low-dimensional space. If a suitable NFM is selected, it can be converted into a nonlinear regression model without changing the form of solving the linear SVR problem, thereby achieving the purpose of enhancing the temporal coding capability of EchoDarwin. Specifically, for a certain function acting on NFM Operation Where D ≥ 2p, Ψ(·) is a nonlinear mapping function, v t is the output value of the nonlinear mapping function, and D is the number of channels after nonlinear mapping. Nonlinearly map it to a higher dimensional space, that is, the processed feature represents a subsequence in the smoothed radar echo subsequence set in the learner Nonlinearly mapping from low-dimensional space to high-dimensional space, the expression of the mapped feature representation learner is as follows:

[0115]

[0116] S325. Solve the expression of the mapped feature representation learner to obtain the optimal solution of the feature representation learner, that is, The optimal solution of the feature representation learner is used as the subsequence of the feature representation learner. The corresponding optimization coefficient;

[0117] Among them, in the specific solution process, the Chi-2 kernel function and the improved Hellinger kernel function PosNeg can be used to define the NFM operation;

[0118] Repeat S322-S325 to obtain the optimization coefficient corresponding to each subsequence in the smoothed radar echo subsequence set;

[0119] In the formula, is the transformed feature representation learner, ||·|| is a norm operator, is the processed feature representation learner, is the L2 regularization term, e is the relaxation factor of the regression error term, l is the penalty factor of the regression error term, [·] ≥0 =max{·;0} is the error loss term that only cares about the regression results that exceed the error relaxation boundary, Φ(V) is the feature representation learner after mapping, is a nonlinear mapping function.

[0120] According to an embodiment of the present invention, in S60 of the present invention, detecting the radar echo sequence to be detected based on the trained detector to obtain the target prediction result includes:

[0121] S61, preprocessing the radar echo sequence to be measured to obtain a preprocessed radar echo subsequence to be measured;

[0122] S62, performing real number conversion and smoothing processing on the preprocessed radar echo subsequence to be measured to obtain a smoothed radar echo subsequence to be measured;

[0123] S63, using a feature representation learner with temporal feature representation to extract features from the smoothed radar echo subsequence to be measured, to obtain features to be measured that maintain a first-order partial order relationship;

[0124] S64, inputting the features to be tested that maintain the first-order partial order relationship into the trained detector to detect the radar echo sequence to be tested, and obtaining the target prediction result.

[0125] Specifically, in S61 of the present invention, the radar echo sequence to be measured is preprocessed to obtain the preprocessed radar echo subsequence to be measured, which includes:

[0126] S611, removing abnormal points from the radar echo sequence to be measured to obtain a radar echo sequence to be measured without abnormal points;

[0127] S612, performing digital pulse compression on the radar echo sequence to be tested without abnormal points to obtain a compressed radar echo sequence to be tested;

[0128] S613, performing finite impulse response filtering on the compressed radar echo sequence to be measured to obtain a filtered radar echo sequence to be measured;

[0129] S614, performing sliding window segmentation on the filtered radar echo sequence to be measured to obtain a preprocessed radar echo subsequence to be measured.

[0130] Among them, in S612 of the present invention, digital pulse compression is performed on the radar echo sequence to be measured without abnormal points, and the compressed radar echo sequence to be measured is obtained, which includes:

[0131] S6121, performing a channel-by-channel Fourier transform on the radar echo sequence to be tested without abnormal points to obtain a frequency domain response curve of a pulse corresponding to each channel;

[0132] S6122. Perform complex restoration on the original reference transmit signal to obtain a restored reference transmit signal;

[0133] S6123, multiplying the frequency domain response curve of each pulse by the restored reference transmission signal bit by bit and performing inverse Fourier transform on a channel by channel basis to obtain a compressed radar echo sequence to be measured;

[0134] In S613 of the present invention, the compressed radar echo sequence to be measured is subjected to finite impulse response filtering to obtain the filtered radar echo sequence to be measured, which includes:

[0135] The compressed radar echo sequence to be measured is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence to be measured:

[0136]

[0137] In the formula, Z p ′(n) is the radar echo sequence to be measured after filtering of the nth range unit of the pth channel, K is the total number of filters, H k,p is the coefficient matrix of the pth channel of the kth filter, Y k ' T (n) is the transpose of the compressed radar echo sequence to be measured of the nth range unit of the kth filter.

[0138] According to an embodiment of the present invention, in order to solve the detection problem caused by the extremely unbalanced positive and negative samples in the low, slow, small and unmanned warning task in the actual situation, in S10 of the present invention, after preprocessing the radar echo sequence set to obtain the preprocessed radar echo subsequence set, the method also includes: upsampling the target sample set in the preprocessed radar echo subsequence set, and downsampling the clutter sample set to obtain the radar echo subsequence set after equalization processing.

[0139] For example, Figure 4 As shown, assuming that the preprocessed radar echo subsequence set has M target echo subsequences, the target sample set Upsampling the target sample set in the preprocessed radar echo subsequence set includes the following steps:

[0140] S1. Vectorize the target sample set in the preprocessed radar echo subsequence set to obtain a vectorized target sample set

[0141] S2, the target sample set after vectorization with replacement Randomly select a target sample from And calculate the randomly selected target sample a nearest neighbor samples of

[0142] S3, from a nearest neighbor samples Randomly select a neighbor sample from Vectorize and obtain the vectorized neighbor samples Among them, γ is a random number between 0 and 1;

[0143] S4. Perform vector slicing restoration on the vectorized neighbor samples to obtain restored neighbor samples

[0144] S5. Restore the nearest neighbor samples As a new target sample, and put it into the generated target sample set In

[0145] S6. Repeat S2-S5 L×M times to obtain a generated target sample set with L×M new target samples. The target sample set will be generated Put it into the target sample set in the preprocessed radar echo subsequence set to obtain the target sample set after equalization processing Right now Thus, upsampling of the target sample set is completed, where L is the synthesis multiple.

[0146] After the above-mentioned sample equalization processing, the clutter sample tilt degree of the model is effectively improved, alleviating the problem of insufficient target sample driving force in radar target detection tasks based on machine learning.

[0147] The present invention also provides a computer device, comprising a memory, a processor, and a target detection program based on radar echo timing dependency stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned methods when executing the target detection program based on radar echo timing dependency.

[0148] In summary, the present invention provides a target detection method and device based on the time-series dependency of radar echoes, by preprocessing the radar echo sequence set to obtain an echo signal with a higher signal-to-noise ratio; the time-series feature representation of the feature representation learner is obtained through the first-order partial order relationship, thereby obtaining a feature representation learner with time-series feature representation, and the feature representation learner is used to extract features and perform subsequent training, thereby realizing target detection using the time-series dependency of radar echoes, and solving the problem that the radar target detection method in the prior art lacks learning of echo feature representation. The present invention can be used in different task fields such as perception, control, and decision-making, and can also be directly applied to radar equipment with various detection systems such as continuous wave and pulse Doppler.

[0149] Parts of the present invention that are not described in detail are well known to those skilled in the art.

[0150] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the devices or elements referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention. The directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0151] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0152] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A target detection method based on radar echo temporal dependency, characterized in that: The method comprises: preprocessing a radar echo sequence set to obtain a preprocessed radar echo subsequence set; Performing real number conversion and smoothing processing on each subsequence in the preprocessed radar echo subsequence set one by one to obtain a smoothed radar echo subsequence set; For each subsequence in the smoothed radar echo subsequence set, the corresponding optimization coefficient of the feature representation learner is obtained based on the first-order partial order relationship; A feature representation learner with an optimized coefficient is used to extract features from the corresponding subsequences in the smoothed radar echo subsequence set one by one, and a feature set that maintains a first-order partial order relationship is obtained; The detector is trained based on the feature set and target identification that maintain the first-order partial order relationship to obtain a trained detector; Based on the trained detector, the radar echo sequence to be tested is detected to obtain the target prediction result.

2. The method according to claim 1, characterized in that For each subsequence in the smoothed radar echo subsequence set, the corresponding optimization coefficient of the feature representation learner obtained based on the first-order partial order relationship includes: The expression of the feature representation learner obtained based on the first-order partial order relationship is as follows: Obtaining a corresponding optimization coefficient of the feature representation learner based on the expression of the feature representation learner; In the formula, is the feature representation learner, is a subsequence in the smoothed radar echo subsequence set, are the first, tth, and Tth elements in the current subsequence respectively, and T is the number of distance units. The energy function for reconstructing the first-order partial order relation for the measurement model, for The metric function of , e is a learnable parameter, is a first-order partial order relation, are the i-th and j-th elements in the current subsequence respectively, They are The measurement function of .

3. The method according to claim 1 or 2, characterized in that: The expression based on the feature representation learner to obtain the corresponding optimization coefficient of the feature representation learner includes: If the metric function is defined as a linear function, the constraint condition of the expression of the feature representation learner is as follows: The constraints are regressed element by element After decoupling, the expression of the transformed feature representation learner is as follows: The expression of the transformed feature representation learner is relaxed and regularized, and the expression of the processed feature representation learner is as follows: The processed feature represents a subsequence in the smoothed radar echo subsequence set in the learner Nonlinearly mapping from low-dimensional space to high-dimensional space, the expression of the mapped feature representation learner is as follows: Solve the expression of the mapped feature representation learner to obtain the optimal solution of the feature representation learner, and use the optimal solution of the feature representation learner as the subsequence of the feature representation learner. The corresponding optimization coefficient; In the formula, is the transformed feature representation learner, ||·|| is a norm operator, is the processed feature representation learner, is the L2 regularization term, e is the relaxation factor of the regression error term, λ is the penalty factor of the regression error term, [·] ≥0 =max{·;0} is the error loss term that only cares about the regression results that exceed the error relaxation boundary, Φ(V) is the feature representation learner after mapping, is a nonlinear mapping function.

4. The method according to claim 1, characterized in that: After preprocessing the radar echo sequence set to obtain the preprocessed radar echo subsequence set, the method further includes: upsampling the target sample set in the preprocessed radar echo subsequence set and downsampling the clutter sample set to obtain the radar echo subsequence set after equalization processing.

5. The method according to claim 4, characterized in that Upsampling the target sample set in the preprocessed radar echo subsequence set includes: S1, vectorizing the target sample set in the preprocessed radar echo subsequence set to obtain a vectorized target sample set; S2, randomly select a target sample from the vectorized target sample set with replacement, and calculate a nearest neighbor samples of the randomly selected target sample; S3, randomly select a neighbor sample from a neighbor samples for vectorization, and obtain the vectorized neighbor sample; S4, performing vector slicing restoration on the vectorized neighbor samples to obtain restored neighbor samples; S5, taking the restored neighbor samples as new target samples and putting them into the generated target sample set; S6. Repeat S2-S5 L×M times to obtain a generated target sample set having L×M new target samples, and put the generated target sample set into the target sample set in the preprocessed radar echo subsequence set to obtain the target sample set after equalization processing, thereby completing the upsampling of the target sample set, wherein L is the synthesis multiple, and M is the number of target echo subsequences in the preprocessed radar echo subsequence set.

6. The method according to claim 1, characterized in that The radar echo sequence set is preprocessed to obtain a preprocessed radar echo subsequence set including: Remove abnormal points from the radar echo sequence set to obtain a radar echo sequence set without abnormal points; Performing digital pulse compression on a radar echo sequence set without abnormal points to obtain a compressed radar echo sequence set; Performing finite impulse response filtering on the compressed radar echo sequence set to obtain a filtered radar echo sequence set; The filtered radar echo sequence set is segmented by a sliding window to obtain a preprocessed radar echo subsequence set.

7. The method according to claim 6, characterized in that Digital pulse compression is performed on the radar echo sequence set without abnormal points, and the compressed radar echo sequence set includes: Perform Fourier transform on the radar echo sequence set without abnormal points channel by channel to obtain the frequency domain response curve of the pulse corresponding to each channel. Performing complex restoration on the original reference transmission signal to obtain a restored reference transmission signal; The frequency domain response curve of each pulse is multiplied bit by bit with the restored reference transmission signal and inverse Fourier transformed channel by channel to obtain a compressed radar echo sequence set; The compressed radar echo sequence set is subjected to finite impulse response filtering, and the filtered radar echo sequence set includes: The compressed radar echo sequence set is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence set: In the formula, Z p (n) is the filtered radar echo sequence of the nth range unit of the pth channel, K is the total number of filters, H k,p is the coefficient matrix of the p-th channel of the k-th filter, is the transpose of the compressed radar echo sequence of the nth range unit of the kth filter.

8. The method according to claim 1, characterized in that Based on the trained detector, the radar echo sequence to be tested is detected, and the target prediction results include: Preprocessing the radar echo sequence to be measured to obtain a preprocessed radar echo subsequence to be measured; The preprocessed radar echo subsequence to be measured is converted into real numbers and smoothed to obtain a smoothed radar echo subsequence to be measured; The feature representation learner with temporal feature representation is used to extract the features of the smoothed radar echo subsequence to be measured, and the features to be measured that maintain the first-order partial order relationship are obtained. The features to be tested that maintain the first-order partial order relationship are input into the trained detector to detect the radar echo sequence to be tested, and the target prediction result is obtained.

9. The method according to claim 8, characterized in that The radar echo sequence to be measured is preprocessed to obtain the radar echo subsequence to be measured after preprocessing, which includes: Remove abnormal points from the radar echo sequence to be tested, and obtain a radar echo sequence to be tested without abnormal points; Performing digital pulse compression on the radar echo sequence to be tested without abnormal points to obtain a compressed radar echo sequence to be tested; Performing finite impulse response filtering on the compressed radar echo sequence to be measured to obtain a filtered radar echo sequence to be measured; The filtered radar echo sequence to be measured is segmented by a sliding window to obtain a preprocessed radar echo subsequence to be measured.

10. The method according to claim 9, characterized in that Digital pulse compression is performed on the radar echo sequence to be tested without abnormal points, and the compressed radar echo sequence to be tested includes: Perform Fourier transform on the radar echo sequence to be tested without abnormal points, and obtain the frequency domain response curve of the pulse corresponding to each channel. Performing complex restoration on the original reference transmission signal to obtain a restored reference transmission signal; The frequency domain response curve of each pulse is multiplied bit by bit with the restored reference transmission signal and inverse Fourier transformed channel by channel to obtain a compressed radar echo sequence to be tested; The compressed radar echo sequence to be measured is subjected to finite impulse response filtering, and the filtered radar echo sequence to be measured includes: The compressed radar echo sequence to be measured is subjected to finite impulse response filtering by the following formula to obtain the filtered radar echo sequence to be measured: In the formula, Z p ′(n) is the radar echo sequence to be measured after filtering of the nth range unit of the pth channel, K is the total number of filters, H k,p is the coefficient matrix of the p-th channel of the k-th filter, is the transpose of the compressed radar echo sequence to be measured of the nth range unit of the kth filter.

11. A computer device, characterized in that: The invention comprises a memory, a processor and a target detection program based on the timing dependency of radar echoes which is stored in the memory and can be run on the processor. When the processor executes the target detection program based on the timing dependency of radar echoes, the method according to any one of claims 1 to 10 is implemented.

Images