Radar distributed target detection method and system based on data dimension reduction

Abstract

The invention belongs to but is not limited to the technical field of target detection, and particularly relates to a radar distributed target detection method and system based on data dimension reduction. And then estimating a noise power mismatch amount, a target echo amplitude vector, a signal steering vector coordinate vector and a noise covariance matrix based on the test data and the auxiliary data after dimension reduction, further calculating a detection statistic, determining a detection threshold according to a false alarm probability, and finally comparing the detection statistic with the detection threshold to determine a noise covariance matrix. The target exists when the detection statistic is greater than the detection threshold, otherwise, the target does not exist. According to the method, the to-be-detected data and the auxiliary data are projected to the signal subspace, so that the data dimension is effectively reduced, a target detection mathematical model is simplified, and reliable detection performance is realized when the auxiliary data volume is limited.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of target detection technology, and in particular relates to a radar distributed target detection method and system based on data dimensionality reduction. Background Art

[0002] Adaptive target detection is one of the basic research directions in the field of radar signal processing. When using high-resolution radar observation, the target can be resolved into several isolated scattering points distributed in multiple distance resolution units to be measured. Ideally, the echoes of these scattering points can be described as the same steering vector weighted by different echo intensities. However, due to factors such as array calibration errors, the assumed steering vector deviates from the true value. In order to cover the deviation of the steering vector to ensure the reliability of the detection results, a distributed model can be used to describe the entire target echo, that is, the steering vector belongs to a known subspace, but the specific coordinates are unknown.

[0003] The radar received signal also contains unknown environmental noise. Auxiliary data containing only noise can be obtained from the adjacent range resolution unit to estimate the noise covariance matrix. Due to noise fluctuations, there is a noise power mismatch between the data to be measured in the observation unit and the auxiliary data in the adjacent unit. This situation is called a partially uniform environment.

[0004] In order to detect distributed targets in partially uniform environments, Olivier Besson et al. designed a generalized adaptive direction detector (GADD) based on the generalized likelihood ratio criterion in 2006. GADD is robust to steering vector deviations. In 2020, Liu Jun et al. derived the same detector based on the generalized likelihood ratio criterion and the Wald criterion under the assumption that the noise covariance matrix has a fully symmetric structure. This detector has a similar structure to GADD and is also robust to steering vector deviations.

[0005] In the context of Gaussian colored noise, the premise for existing detectors to ensure good detection performance is to estimate the noise covariance matrix with high accuracy, which requires obtaining sufficient auxiliary data from neighboring units, usually requiring the amount of auxiliary data to be more than twice the dimension of the data model. However, the actual environment is continuously fluctuating, and the number of neighboring units with the same noise characteristics as the unit to be tested is limited. At this time, existing detectors will face the risk of drastic performance deterioration or even failure. Therefore, effective target detection under the condition of limited auxiliary data is a serious problem that needs to be solved urgently.

[0006] In view of the above analysis, the technical problem that urgently needs to be solved in the prior art is: how to perform effective target detection when the amount of auxiliary data is limited. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention provides a radar distributed target detection method and system based on data dimensionality reduction, which aims to improve the detection performance of radar distributed targets when the detector has limited auxiliary data in a partially uniform environment. The present invention projects the data to be measured and the auxiliary data into the signal subspace to achieve data dimensionality reduction, which simplifies the target detection model and reduces the dimension of the noise covariance matrix to be estimated, effectively reduces the amount of auxiliary data required to estimate the noise covariance matrix, and ensures reliable detection performance when the amount of auxiliary data is limited.

[0008] The present invention is implemented as follows: a radar distributed target detection method based on data dimensionality reduction, comprising:

[0009] S1: Get radar in K p The data to be tested is sampled in the continuous units to be tested. The signal subspace to which the steering vector of the unit to be detected belongs is represented by the column full rank matrix Zhang Cheng, the coordinate vector of the steering vector in the signal subspace is Under the target assumption H1, the target signal amplitude vector is Where M is the number of radar channels or the data model dimension, Q is the signal subspace dimension, represents a complex matrix of dimension m×n, (·) * denotes conjugation;

[0010] S2: Get the K of the radar from the unit under test s Auxiliary data sampled in adjacent cells

[0011] S3: The data to be tested X p and auxiliary data X s are projected into the signal subspace for data dimension reduction, and new test data are obtained respectively. and new auxiliary data

[0012] S4: Under the target hypothesis H1, according to the new test data Y p and the new auxiliary data Y s The joint probability density function f1(Y p ,Y s ) and the signal subspace matrix H, solve the estimated value of the power mismatch γ of the Gaussian colored noise in the measured data and the auxiliary data Solve the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H Estimated value of Solve the conjugate amplitude vector of the target signal in the measured data Estimated value of Solve the covariance matrix of the new Gaussian colored noise Estimated value of

[0013] S5: Conjugate the target signal to the estimated amplitude vector Column vectorization and storage in the relevant parameter vector The estimated value of the Gaussian colored noise power mismatch between the measured data and the auxiliary data is The estimated value of the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H New covariance matrix estimate of Gaussian colored noise Vectorize the columns and save them in the redundant parameter vector In the above example, the relevant parameter vector θ r and the redundant parameter vector θ s The average columns are vectorized and stored in the parameter vector middle;

[0014] S6: Calculate the detection statistic t based on the Wald criterion according to the parameter vector θ Wald ;

[0015] S7: According to the false alarm probability P fa Determine the detection threshold T Wald ;

[0016] S8: Comparison of the test statistic t based on the Wald criterion Wald and the detection threshold T Wald The size relationship is used to determine whether there is a target signal of interest in the data to be tested.

[0017] Further, in step S3, the test data X p and auxiliary data X s All are projected into the signal subspace for data dimension reduction to obtain the new test data Y p and the new auxiliary data Y s They are: Y p =(H H H) -1 H H X p and Y s =(H H H) -1 H H X s ,in(·) G represents the conjugate transpose of the matrix, (·) --1 Represents the inverse of a reversible matrix.

[0018] Further, the estimated value of the Gaussian colored noise power mismatch γ between the measured data and the auxiliary data in step S4 is for: It is a multivariate equation The only positive solution of , where N = min{K p ,Q},ε1≥ε2≥...≥ε N is the new test data Y p The Hermitian inner product matrix after quasi-whitening The non-zero eigenvalues ​​of is the new auxiliary data Y s The sampling covariance matrix of .

[0019] Further, the estimated value of the coordinate vector p of the target signal in the test data in the signal subspace spanned by the matrix H in step S4 is for: yes The eigenvector corresponding to the maximum eigenvalue of Indicates that the dimension is K s The unit matrix of the target signal; the estimated value of the conjugate amplitude vector α of the target signal in the measured data for:

[0020] Further, the estimated value of the covariance matrix ∑ of the new Gaussian colored noise in step S4 is for:

[0021] Furthermore, the relevant parameter vector θ in step S5 r for: The redundant parameter vector θ s for: Where vec(·) means to vectorize the matrix columns, (·) T represents the transpose; the parameter vector θ is:

[0022] Furthermore, the detection statistic t based on the Wald criterion in step S6 Wald The expression is: in i,j∈{r,s}, E[·] means to find the expectation, ln(·) means the natural logarithm function, f1(Y p ,Y s ) is the new test data Y p and the new auxiliary data Y s Joint probability density function under the target hypothesis H1.

[0023] Further, in step S7, according to the false alarm probability P fa Determine the detection threshold T Wald The specific steps are:

[0024] S71: Under the targetless hypothesis H0, calculate the detection statistic t based on the Wald criterion Wald ;

[0025] S72: Execute M on step S71 c The Monte Carlo simulation is performed to calculate the detection statistic t based on the Wald criterion. Wald Store in column vector Among them represents a real matrix of dimension m×n;

[0026] S73: vector V Wald The elements in V are rearranged in ascending order. Wald Middle The element is the false alarm probability P fa The corresponding detection threshold T Wald ,in Indicates rounding down.

[0027] Furthermore, in step S8, the detection statistic t based on the Wald criterion is compared. Wald and the detection threshold T Wald The specific process of judging whether there is a target signal of interest in the data to be tested is as follows: Wald >T Wald , then the target signal of interest exists; if t Wald ≤T Wald , then the target signal of interest does not exist.

[0028] Another object of the present invention is to provide a radar distributed target detection system based on data dimensionality reduction for realizing the radar distributed target detection method based on data dimensionality reduction, comprising:

[0029] Data acquisition module: used to obtain the data to be tested and auxiliary data;

[0030] Parameter acquisition module: used to obtain the subspace column full rank matrix to which the steering vector of the unit to be detected belongs and the required false alarm probability;

[0031] Data dimension reduction module: used to project the test data and auxiliary data into the signal subspace for data dimension reduction, so as to obtain new test data and new auxiliary data respectively;

[0032] Parameter calculation module: used to calculate the sampling covariance matrix of the new auxiliary data using the new data to be measured, the new auxiliary data and the signal subspace matrix, calculate the power mismatch of the Gaussian colored noise in the data to be measured and the auxiliary data, calculate the coordinate vector of the target signal in the data to be measured in the subspace spanned by the signal subspace matrix, calculate the amplitude vector of the target signal in the data to be measured, and calculate the covariance matrix of the new Gaussian colored noise;

[0033] Detection statistic calculation module: used to calculate the detection statistic based on the Wald criterion using the results output by the parameter calculation module;

[0034] Detection threshold calculation module: used to calculate the detection threshold according to the false alarm probability;

[0035] Target detection output module: used to compare the detection statistics and the detection threshold, and output the decision result of whether there is a target signal of interest in the data to be tested.

[0036] Another object of the present invention is to provide a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the radar distributed target detection method based on data dimensionality reduction.

[0037] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the radar distributed target detection method based on data dimensionality reduction.

[0038] Another object of the present invention is to provide an information data processing terminal, which includes the radar distributed target detection system based on data dimensionality reduction.

[0039] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0040] First, the present invention achieves data dimensionality reduction by projecting the data to be measured and the auxiliary data into the signal subspace, which greatly simplifies the mathematical model of target detection. The present invention reduces the dimension of the noise covariance matrix to be estimated and reduces the detector's demand for the amount of auxiliary data. The present invention has a smaller calculation complexity of the detection statistic and achieves higher distributed target detection performance when the amount of auxiliary data is limited.

[0041] Second, the technical solution of the present invention solves the technical problems that people have been eager to solve but have never been able to solve successfully:

[0042] ① In reality, factors such as array calibration errors, radar channel or array element consistency problems, radar antenna beam pointing deviations, etc. can cause the target steering vector to deviate in different directions. In this case, if the target steering vector is assumed to be accurately known, the detection results of conventional detectors will have reliability problems. By covering the deviation range of the steering vector through the signal subspace model, this modeling method has slightly deteriorated performance when the target echo steering vector has no deviation, but it can ensure the reliability of the detection results when it is uncertain whether the steering vector has deviations and to what extent.

[0043] ② When performing target detection, it is necessary to sample sufficient auxiliary data from the neighboring units of the unit to be observed to estimate the noise covariance matrix in the unit to be observed. However, the current detection environment is becoming more and more complex and changeable, and the noise characteristics in different distance resolution units fluctuate greatly. In order to ensure the accuracy of the noise covariance matrix estimation results in the unit to be observed, auxiliary data can only be sampled from several units adjacent to the unit to be observed. At this time, there is a problem of insufficient auxiliary data and the performance of existing detectors deteriorates or even fails. Therefore, how to perform effective and reliable target detection under the condition of limited auxiliary data is a very necessary, meaningful and urgent problem to be solved. The present invention reduces the dimension of the detection problem and the dimension of the noise covariance matrix to be estimated by projecting the data into the target signal subspace as a preprocessing method, thereby effectively reducing the detector's demand for auxiliary data. In addition, data projection also reduces the computational complexity of the detection statistic, which is conducive to real-time detection.

[0044] The technical solution of the present invention overcomes technical prejudice: the existing technical solutions are all based on specific criteria (such as generalized likelihood ratio criterion, Rao criterion, Wald criterion, etc.) to design detectors. This design process is carried out in the entire observation space, that is, the dimension of the processed data is equal to the number of radar channels. However, the core of target detection is to determine whether there is a target signal of interest in the signal subspace, so detection can be performed only in the signal subspace. The technical solution of the present invention is to introduce a data preprocessing process, that is, first project the data into the signal subspace to achieve data dimensionality reduction, and then design the detector based on specific criteria. Since the dimension of the signal subspace is smaller than or even much smaller than the number of radar channels in practice, the complexity of the detector design is significantly reduced after the data dimensionality reduction. In addition, projecting the data into the signal subspace can eliminate the need to consider the influence of noise in the orthogonal signal subspace, which is beneficial to improving the detection performance.

[0045] Third, the technical solution of the present invention solves several key problems in the prior art in industrial applications and has achieved significant technical progress, which is specifically reflected in the following aspects:

[0046] 1. Data dimensionality reduction improves processing efficiency: When facing complex environments, existing radar target detection systems often consume a lot of computing resources to process large-scale data, and the detection accuracy is greatly affected by noise. The present invention introduces a data dimensionality reduction module to project the data to be tested and the auxiliary data into the signal subspace, effectively reducing the dimension of the data and the computational complexity. This not only improves the processing efficiency of the system, but also reduces the resource usage while ensuring accuracy.

[0047] 2. Suppress the noise outside the signal subspace: The traditional radar detection method is carried out in the entire observation space filled with noise. The noise in the orthogonal signal subspace will lead to false detection and missed detection. The present invention eliminates the influence of noise in the orthogonal signal subspace by projecting data into the signal subspace, thereby improving the accuracy of detection, especially in complex environments.

[0048] 3. Reduce false alarm probability and improve detection sensitivity: Traditional radar detection systems often find it difficult to maintain high detection sensitivity in complex backgrounds while maintaining a low false alarm probability. The present invention can significantly improve the detection sensitivity of weak target signals while maintaining a low false alarm probability by introducing the projection of the signal subspace matrix and accurate detection statistic calculation, and is particularly suitable for application in the field of radar distributed target detection.

[0049] 4. Adapt to partially uniform environment: The existing radar detection system has the problem of reduced detection performance when dealing with partially uniform environment. To address this problem, the present invention introduces an unknown mismatch to characterize the difference in noise fluctuations between the unit under test and the adjacent units, and incorporates the estimation of the mismatch into the design process of the detector, thereby ensuring that the present invention is insensitive to the difference in noise fluctuations between the unit under test and the adjacent units, effectively coping with the detection challenges in partially uniform environments, and greatly improving the performance in practical applications.

[0050] 5. Broad industrial application prospects: The present invention solves the difficult problem of large data processing volume and difficulty in balancing detection accuracy and efficiency in the field of radar target detection. The technical solution can be widely used in multiple industries such as radar, communications, and national defense, especially in distributed radar systems and target detection in complex environments. It has significant application value and market prospects.

[0051] In summary, the present invention has achieved significant technological progress while solving the problems of the prior art through technical means such as data dimensionality reduction, suppression of noise outside the signal subspace, and detection statistic calculation, and has strong potential for industrial application. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of a radar distributed target detection method based on data dimensionality reduction provided by an embodiment of the present invention;

[0053] Figure 2 It is a structural diagram of a radar distributed target detection system based on data dimensionality reduction provided by an embodiment of the present invention;

[0054] Figure 3 is a comparison diagram of detection probabilities of the present invention and the prior art solution under different auxiliary data amounts provided by an embodiment of the present invention;

[0055] Figure 4 It is a comparison diagram of detection probabilities under different signal-to-noise ratios provided by an embodiment of the present invention and a prior art solution. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] The present invention proposes a radar distributed target detection method based on data dimensionality reduction, which focuses on solving technical problems such as low detection accuracy, high false alarm rate, and heavy computational burden in traditional radar target detection methods under complex environments, and improves detection efficiency and accuracy through data dimensionality reduction and parameter optimization, which has significant technical progress.

[0058] First, traditional radar target detection has the problem of high computational complexity and reduced detection accuracy when facing high-dimensional data and complex environments. The present invention projects high-dimensional radar data into the signal subspace through data dimensionality reduction technology, thereby effectively reducing the data dimension and significantly reducing the computational burden. At the same time, the dimensionality reduction operation not only retains the main information of the target signal, but also reduces the influence of background noise and interference, so that the system can process a large amount of test data more efficiently.

[0059] Secondly, in terms of the balance between false alarm rate and detection rate, traditional methods are often limited by the influence of noise interference and it is difficult to obtain a low false alarm rate. Based on the Wald criterion, the present invention combines the extraction of relevant parameters and redundant parameters, and dynamically adjusts the detection threshold through the optimized detection statistics, thereby ensuring that the false alarm rate is controlled within a certain range while maintaining high sensitivity detection of the target signal. This improvement not only improves the robustness of the detection, but also further enhances the adaptability and accuracy of the system.

[0060] In addition, in order to solve the problem that traditional methods are difficult to effectively estimate noise mismatch and signal strength, the present invention uses a joint probability density function in the reduced-dimensional data space to accurately estimate parameters such as the power mismatch of Gaussian colored noise, the amplitude vector of the target signal, and the covariance matrix, thereby reducing the system's dependence on preset parameters. Using these parameters, more accurate detection statistics are constructed, effectively improving the stability and adaptability of the system in different environments.

[0061] The detection method of the present invention not only effectively improves the detection efficiency of the radar target detection system in complex scenarios and reduces the false alarm rate, but also significantly improves the detection accuracy and robustness, providing a more reliable and intelligent solution for the application of radar target detection in military, aviation, transportation and other industries, showing significant technological progress and industrial application prospects.

[0062] The following are two specific application examples of the radar distributed target detection method based on the present invention in actual industry:

[0063] Example 1: Object Detection System in Autonomous Driving

[0064] In the autonomous driving industry, vehicles need to accurately identify and detect surrounding moving and stationary targets, such as pedestrians, other vehicles, and obstacles. The present invention can reduce the impact of noise in complex urban environments and achieve more accurate target detection through data dimension reduction and noise optimization. By using the continuous scanning data of the radar and adjusting the detection threshold through the optimized detection algorithm and Wald criterion, the system can efficiently detect various targets while maintaining a low false alarm rate and a high response speed, thereby ensuring the safety and reliability of the autonomous driving system.

[0065] Example 2: Airport Ground Monitoring Radar System

[0066] The airport's ground surveillance radar needs to monitor the dynamic conditions of areas such as runways, taxiways and aprons in real time to ensure the safe operation of aircraft and ground vehicles. Traditional radar detection is easily affected by complex ground echo interference, while the method of the present invention can effectively filter out ground clutter through signal subspace dimensionality reduction and noise estimation optimization. The system can accurately detect ground activities under different weather and light conditions, ensuring efficient monitoring around the clock and improving the overall operational safety and efficiency of the airport.

[0067] It is known that a radar detection system has M channels. p The test data sampled in the continuous units to be tested is in is the test data in the kth unit to be tested, k = 1, 2, ..., K p , represents a complex matrix of dimension m×n. Under the untargeted hypothesis test H0, the test data X p Contains only Gaussian colored noise N p All column vectors of are independent and identically distributed and have a mean of zero, and the covariance matrix is The Gaussian distribution of the scalar γ represents the noise power mismatch between the unit to be detected and its neighboring units. Under the target hypothesis detection H1, the test data X p Contains Gaussian colored noise N p and distributed target echo sα H ,in is the target orientation vector, is the target echo conjugate amplitude vector, is the target echo amplitude in the kth unit to be detected, k = 1, 2, ..., K p , (·) T represents transpose, (·) * Influenced by factors such as radar channel calibration error, the steering vector s is not precisely known but belongs to a known column full rank matrix The signal subspace spanned is s = Hp, where is an unknown coordinate vector, which indicates the position of the steering vector s in the signal subspace. In order to estimate the noise covariance matrix γR in the data to be tested, we can get the K s The auxiliary data sampled in the adjacent cells is Auxiliary DataX s Contains only Gaussian colored noise Represents the auxiliary data in the lth neighboring unit, l = 1, 2, ..., K s . Auxiliary data X s =N s All column vectors in are independent and identically distributed and follow a Gaussian distribution with a mean of zero and a covariance matrix of R. Therefore, the binary hypothesis testing problem for distributed targets in a partially uniform environment can be expressed as:

[0068] The purpose of the present invention is to improve the detection performance of radar distributed targets in a partially uniform environment when the amount of auxiliary data is limited. To this end, this embodiment provides a radar distributed target detection method based on data dimensionality reduction in a partially uniform environment, such as Figure 1 As shown, the steps of the method include:

[0069] S1: Get radar in K p The data to be tested is sampled in the continuous units to be tested. The signal subspace to which the steering vector of the unit to be detected belongs is represented by the column full rank matrix Zhang Cheng, the coordinate vector of the steering vector in the signal subspace is Under the target assumption H1, the target signal amplitude vector is Where M is the number of radar channels or the data model dimension, Q is the signal subspace dimension, represents a complex matrix of dimension m×n, (·) * denotes conjugation;

[0070] S2: Get the K of the radar from the unit under test s Auxiliary data sampled in adjacent cells

[0071] S3: The data to be tested X p and auxiliary data X s are projected into the signal subspace for data dimension reduction, and new test data are obtained respectively. and new auxiliary data

[0072] S4: Under the target hypothesis H1, according to the new test data Y p and the new auxiliary data Y s The joint probability density function f1(Y p ,Y s ) and the signal subspace matrix H, solve the estimated value of the power mismatch γ of the Gaussian colored noise in the measured data and the auxiliary data Solve the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H Estimated value of Solve the conjugate amplitude vector of the target signal in the measured data Estimated value of Solve the covariance matrix of the new Gaussian colored noise Estimated value of

[0073] S5: Conjugate the target signal to the estimated amplitude vector Column vectorization and storage in the relevant parameter vector The estimated value of the Gaussian colored noise power mismatch between the measured data and the auxiliary data is The estimated value of the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H New covariance matrix estimate of Gaussian colored noise Vectorize the columns and save them in the redundant parameter vector In the above example, the relevant parameter vector θ r and the redundant parameter vector θ s The average columns are vectorized and stored in the parameter vector middle;

[0074] S6: Calculate the detection statistic t based on the Wald criterion according to the parameter vector θWald ;

[0075] S7: According to the false alarm probability P fa Determine the detection threshold T Wald ;

[0076] S8: Comparison of the test statistic t based on the Wald criterion Wald and the detection threshold T Wald The size relationship is used to determine whether there is a target signal of interest in the data to be tested.

[0077] In step S3, the data to be tested X p and auxiliary data X s All are projected into the signal subspace for data dimension reduction to obtain the new test data Y p and the new auxiliary data Y s They are: Y p =(H H H) -1 H H X p and Y s =(H H H) -1 H H X s ,in(·) H represents the conjugate transpose of the matrix, (·) -1 Represents the inverse of a reversible matrix.

[0078] The estimated value of the Gaussian colored noise power mismatch γ between the measured data and the auxiliary data in step S4 for: It is a multivariate equation The only positive solution of , where N = min{K p ,Q},ε1≥ε2≥…≥ε N is the new test data Y p The Hermitian inner product matrix after quasi-whitening The non-zero eigenvalues ​​of is the new auxiliary data Y s The sampling covariance matrix of .

[0079] The estimated value of the coordinate vector p of the target signal in the data to be measured in step S4 in the signal subspace spanned by the matrix H for: yes The eigenvector corresponding to the maximum eigenvalue of Indicates that the dimension is K s The unit matrix of the target signal; the estimated value of the conjugate amplitude vector α of the target signal in the measured data for:

[0080] The estimated value of the covariance matrix Σ of the new Gaussian colored noise in step S4 is for:

[0081] The relevant parameter vector θ in step S5 r for: The redundant parameter vector θ s for: Where vec(·) means to vectorize the matrix columns, (·) T represents the transpose; the parameter vector θ is:

[0082] The detection statistic t based on the Wald criterion in step S6 Wald The expression is: in i,j∈{r,s}, E[·] means to find the expectation, ln(·) means the natural logarithm function, f1(Y p ,Y s ) is the new test data Y p and the new auxiliary data Y s Joint probability density function under the target hypothesis H1.

[0083] In step S7, according to the false alarm probability P fa Determine the detection threshold T Wald The specific steps are:

[0084] S71: Under the targetless hypothesis H0, calculate the detection statistic t based on the Wald criterion Wald ;

[0085] S72: Execute M on step S71 c The Monte Carlo simulation is performed to calculate the detection statistic t based on the Wald criterion. Wald Store in column vector Among them represents a real matrix of dimension m×n;

[0086] S73: vector V Wald The elements in V are rearranged in ascending order. Wald Middle The element is the false alarm probability P fa The corresponding detection threshold T Wald ,in Indicates rounding down.

[0087] In step S8, the detection statistic t based on the Wald criterion is compared.Wald and the detection threshold T Wald The specific process of judging whether there is a target signal of interest in the data to be tested is as follows: Wald >T Wald , then the target signal of interest exists; if t Wald ≤T Wald , then the target signal of interest does not exist.

[0088] See also Figure 2 As shown, the present invention provides a radar distributed target detection system based on data dimension reduction in a partially uniform environment, comprising:

[0089] Data acquisition module: used to obtain the data to be tested and auxiliary data;

[0090] Parameter acquisition module: used to obtain the subspace column full rank matrix to which the steering vector of the unit to be detected belongs and the required false alarm probability;

[0091] Data dimension reduction module: used to project the test data and auxiliary data into the signal subspace for data dimension reduction, so as to obtain new test data and new auxiliary data respectively;

[0092] Parameter calculation module: used to calculate the sampling covariance matrix of the new auxiliary data using the new data to be measured, the new auxiliary data and the signal subspace matrix, calculate the power mismatch of the Gaussian colored noise in the data to be measured and the auxiliary data, calculate the coordinate vector of the target signal in the data to be measured in the subspace spanned by the signal subspace matrix, calculate the amplitude vector of the target signal in the data to be measured, and calculate the covariance matrix of the new Gaussian colored noise;

[0093] Detection statistic calculation module: used to calculate the detection statistic based on the Wald criterion using the results output by the parameter calculation module;

[0094] Detection threshold calculation module: used to calculate the detection threshold according to the false alarm probability;

[0095] Target detection output module: used to compare the detection statistics and the detection threshold, and output the decision result of whether there is a target signal of interest in the data to be tested.

[0096] The radar distributed target detection system based on data dimension reduction provided by the present invention first obtains the data to be measured and auxiliary data through the data acquisition module. These data include the original radar signal and related environment or background information, which are used for subsequent target detection processing.

[0097] Secondly, the column full-rank matrix of the signal subspace and the system preset false alarm probability are obtained from the parameter acquisition module to ensure that the statistical confidence of the detection meets the design requirements, thereby improving the detection accuracy of the system.

[0098] Next, the data dimension reduction module realizes data dimension reduction by projecting the test data and auxiliary data into the signal subspace. After dimension reduction, the system generates new test data and new auxiliary data. This dimension reduction process can remove excess noise and redundant information, improve the ability to distinguish target signals from background noise, and reduce computational complexity.

[0099] Based on the reduced-dimensional data, the parameter calculation module starts to calculate various key parameters. First, it calculates the sampling covariance matrix of the new auxiliary data and derives the power mismatch of the Gaussian colored noise based on these matrices. At the same time, the module calculates the coordinate vector and amplitude vector of the target signal in the data to be tested in the signal subspace. These calculations help to accurately characterize the characteristics of the target signal.

[0100] Subsequently, the detection statistic calculation module calculates the detection statistic based on the above calculation results using the Wald criterion. The Wald criterion is an effective detection method that can quickly determine whether there is a target signal in the data to be tested. Through the calculation of this statistic, the system can generate a series of valid values ​​for identifying the target.

[0101] Finally, the detection threshold calculation module calculates the detection threshold according to the preset false alarm probability and compares it with the detection statistic. Based on this comparison result, the target detection output module determines whether there is a target signal of interest in the data to be tested and outputs the final detection decision. This result provides the radar system with reliable target detection capabilities, especially in partially uniform environments.

[0102] The effect of the present invention on detecting distributed targets in a partially uniform environment is further illustrated below based on simulation experiments.

[0103] Figure 3 The comparison diagram of detection probability between the technical solution of the present invention and the existing generalized adaptive direction detector (GADD) based on the generalized likelihood ratio criterion under different auxiliary data amounts is shown, where the simulation conditions are: false alarm probability P fa =0.001, the number of radar detection system channels M = 12, the signal subspace dimension Q = 6, the amount of data to be measured K p =10, auxiliary data volume K s From Q-2=4 to 3M=36, the signal-to-noise ratio is α H α×p H H H (γR) -1 Hp = 22dB, where γR is the noise covariance matrix in the data to be measured. Figure 3 It can be seen that GADD needs the amount of auxiliary data Ks It is only valid when it is greater than the observation dimension M and when K s ≥2M can guarantee a detection probability of more than 80%; compared with GADD, the method proposed in this invention has a lower auxiliary data volume K s It is effective when it is greater than the signal subspace dimension Q, and when M≤K s When the detection probability is ≤2M, the detection probability is increased by more than 20%. When the detection probability is 80%, the method proposed in the present invention has a significant impact on the auxiliary data volume K. s The demand has decreased When the amount of auxiliary data is sufficient, such as K s When ≥3M, the performance of GADD is consistent with that of the method proposed in the present invention, and both achieve 100% detection probability.

[0104] Figure 4 The comparison diagram of detection probability between the technical solution of the present invention and the existing GADD under different signal-to-noise ratios is shown, where the simulation conditions are: false alarm probability P fa =0.001, the number of radar detection system channels M = 12, the signal subspace dimension Q = 6, the amount of data to be measured K p =10, auxiliary data volume K s =2M=24, the signal-to-noise ratio increases from 10dB to 30dB at equal intervals. Figure 4 It can be seen from the figure that under different signal-to-noise ratios, the method proposed in the present invention has an advantage over GADD in terms of detection probability. Equivalently, when achieving the same detection probability, the method proposed in the present invention requires a lower signal-to-noise ratio: Figure 4 As shown, the signal-to-noise ratio required by the method of the present invention is reduced by about 1.7 dB to achieve an 80% detection probability. When the signal-to-noise ratio is large, such as when the signal-to-noise ratio is ≥ 25 dB, the performance of GADD and the method of the present invention is consistent, and both can achieve a 100% detection probability.

[0105] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of a radar distributed target detection method based on data dimensionality reduction.

[0106] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a radar distributed target detection method based on data dimensionality reduction.

[0107] An application embodiment of the present invention provides an information data processing terminal, which includes a radar distributed target detection system based on data dimensionality reduction.

[0108] The present invention can be applied to all non-contact, high-precision target detection fields, such as

[0109] ① Military field: Currently all military radars need to detect targets and then perform early warning, tracking, navigation, fire control and other tasks according to different tactical requirements.

[0110] ② Civil aviation field: The present invention can be used to detect aircraft for air traffic control, and can also detect birds, drones and other unidentified flying objects around airports for early warning.

[0111] ③ Traffic field: The present invention can be used to detect vehicles and pedestrians, and then monitor traffic flow, traffic accidents and congestion, etc. The present invention can also be used for intelligent driving, helping the intelligent driving system to promptly detect vehicles and pedestrians around it.

[0112] ④Medical field: The present invention can be used to detect abnormal tissues such as tumors and nodules in the human body.

[0113] The present invention can also be used for meteorological monitoring to predict weather, for environmental protection to monitor forest fires, for industrial production positioning robots, etc.

[0114] In summary, the target detection technology involved in the present invention plays an important role in many fields, and the application field of this technology will continue to expand and deepen.

[0115] The technical effect obtained by the embodiment of the present invention is obtained through Monte Carlo simulation. Whether it is the prior art or the technology of the present invention, the detection process involves implicit operations such as finding eigenvalues ​​and eigenvectors of the matrix, so it is impossible to draw a conclusion on which is better through theoretical analysis. Monte Carlo simulation is a numerical calculation method based on probability and statistical theory. In a statistical sense, Monte Carlo simulation is equivalent to theoretical analysis and is also a commonly used numerical analysis method in the industry. Therefore, independent and repeated Monte Carlo simulations can be used to simulate complex detection processes and evaluate the detection probabilities of the prior art and the technology of the present invention under the same environment.

[0116] Figure 3 and Figure 4 The curve shows the comparison results obtained through Monte Carlo simulation, which is true and reliable, and can clearly demonstrate the performance advantages and broader application range of the technology of the present invention compared with the prior art.

[0117] The following are two industrial application examples of radar distributed target detection methods based on data dimensionality reduction:

[0118] 1. Application in marine ship monitoring system

[0119] In the marine environment, radar is widely used to monitor and track ships. However, due to the presence of sea surface waves and clutter signals, traditional radar detection systems often face problems such as signal-noise confusion and high false alarm probability. Using the method of the present invention, the original radar echo can be reduced to the signal subspace through data dimension reduction to remove redundant information. On this basis, the Wald criterion is used to accurately detect the target of interest (such as a ship). By reducing the influence of Gaussian colored noise, the system can improve the sensitivity to weak signals of ships and reduce the false alarm probability in complex sea conditions.

[0120] Actual effect: In a certain sea area application, the radar distributed target detection system effectively reduced the false alarm probability from 8% of the original system to 1.5%, and the accuracy of ship detection increased by 15%. The system is particularly suitable for marine monitoring, maritime search and rescue, and national defense security.

[0121] 2. Application in airport runway foreign object detection system

[0122] In large airports, foreign object detection (FOD) on the runway is crucial. Traditional detection systems are often affected by environmental noise and weather conditions, resulting in low detection accuracy. The radar distributed target detection method of the present invention can effectively cope with these challenges. It reduces the dimension of the detection data through data dimensionality reduction technology, filters out the background noise in the complex orthogonal signal subspace, and improves the detection accuracy of the system for foreign object signals by estimating the power mismatch of Gaussian colored noise. In addition, the detection statistic is calculated using the Wald criterion to achieve efficient target detection and false alarm control.

[0123] Actual effect: In a test application at an international airport, the system successfully detected foreign objects with a diameter of less than 2 cm, with a detection accuracy of about 20% higher than that of traditional systems, and the false alarm probability was significantly reduced to less than 0.8%. The system has broad application prospects in the fields of aviation safety, transportation facility management, etc.

[0124] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0125] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A radar distributed target detection method based on data dimensionality reduction, characterized in that: include: S1: Get radar in K p The data to be tested is sampled in the continuous units to be tested. The signal subspace to which the steering vector of the unit to be detected belongs is represented by the column full rank matrix Zhang Cheng, the coordinate vector of the steering vector in the signal subspace is Under the target assumption H1, the target signal amplitude vector is Where M is the number of radar channels or the data model dimension, Q is the signal subspace dimension, represents a complex matrix of dimension m×n, (·) * denotes conjugation; S2: Get the K of the radar from the unit under test s Auxiliary data sampled in neighboring cells S3: The data to be tested X p and auxiliary data X s are projected into the signal subspace for data dimension reduction, and new test data are obtained respectively. and new auxiliary data S4: Under the target hypothesis H1, according to the new test data Y p and the new auxiliary data Y s The joint probability density function f1(Y p ,Y s ) and the signal subspace matrix H, solve the estimated value of the power mismatch γ of the Gaussian colored noise in the measured data and the auxiliary data Solve the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H Estimated value of Solve the conjugate amplitude vector of the target signal in the measured data Estimated value of Solve the covariance matrix of the new Gaussian colored noise Estimated value of S5: Conjugate the target signal to the estimated amplitude vector Column vectorization and storage in the relevant parameter vector The estimated value of the Gaussian colored noise power mismatch between the measured data and the auxiliary data is The estimated value of the coordinate vector of the target signal in the measured data in the signal subspace spanned by the matrix H New covariance matrix estimate of Gaussian colored noise Vectorize the columns and save them in the redundant parameter vector In the above example, the relevant parameter vector θ r and the redundant parameter vector θ s The average columns are vectorized and stored in the parameter vector middle; S6: Calculate the detection statistic t based on the Wald criterion according to the parameter vector θ Wald ; S7: According to the false alarm probability P fa Determine the detection threshold T Wald ; S8: Comparison of the test statistic t based on the Wald criterion Wald and the detection threshold T Wald The size relationship is used to determine whether there is a target signal of interest in the data to be tested.

2. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: In step S3, the data to be tested X p and auxiliary data X s All are projected into the signal subspace for data dimension reduction to obtain the new test data Y p and the new auxiliary data Y s They are: Y p =(H H H) -1 H H X p and Y s =(H H H) -1 H H X s ,in(·) H represents the conjugate transpose of the matrix, (·) -1 Represents the inverse of a reversible matrix.

3. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: The estimated value of the Gaussian colored noise power mismatch γ between the measured data and the auxiliary data in step S4 for: It is a multivariate equation The only positive solution of , where N = min{K p ,Q},ε1≥ε2≥…≥ε N is the new test data Y p The Hermitian inner product matrix after quasi-whitening The non-zero eigenvalues ​​of is the new auxiliary data Y s The sampling covariance matrix of .

4. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: The estimated value of the coordinate vector p of the target signal in the data to be tested in step S4 in the signal subspace spanned by the matrix H for: yes The eigenvector corresponding to the maximum eigenvalue of Indicates that the dimension is K s The unit matrix of the target signal; the estimated value of the conjugate amplitude vector α of the target signal in the measured data for:

5. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: The estimated value of the covariance matrix ∑ of the new Gaussian colored noise in step S4 is for:

6. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: The relevant parameter vector θ in step S5 r for: The redundant parameter vector θ s for: Where vec(·) means to vectorize the matrix columns, (·) T represents the transpose; the parameter vector θ is:

7. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: The detection statistic t based on the Wald criterion in step S6 Wald The expression is: in i,j∈{r,s}, E[·] means to find the expectation, ln(·) means the natural logarithm function, f1(Y p ,Y s ) is the new test data Y p and the new auxiliary data Y s Joint probability density function under the target hypothesis H1.

8. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: In step S7, according to the false alarm probability P fa Determine the detection threshold T Wald The specific steps are: S71: Under the targetless hypothesis H0, calculate the detection statistic t based on the Wald criterion Wald ; S72: Execute M on step S71 c The Monte Carlo simulation is performed to calculate the detection statistic t based on the Wald criterion. Wald Store in column vector Among them represents a real matrix of dimension m×n; S73: vector V Wald The elements in V are rearranged in ascending order. Wald Middle The element is the false alarm probability P fa The corresponding detection threshold T Wald ,in Indicates rounding down.

9. The radar distributed target detection method based on data dimension reduction as claimed in claim 1, characterized in that: In step S8, the detection statistic t based on the Wald criterion is compared. Wald and the detection threshold T Wald The specific process of judging whether there is a target signal of interest in the data to be tested is as follows: Wald >T Wald , then the target signal of interest exists; if t Wald ≤T Wald , then the target signal of interest does not exist.

10. A radar distributed target detection system based on data dimensionality reduction for implementing the radar distributed target detection method based on data dimensionality reduction as claimed in any one of claims 1 to 9, characterized in that: include: Data acquisition module: used to obtain the data to be tested and auxiliary data; Parameter acquisition module: used to obtain the subspace column full rank matrix to which the steering vector of the unit to be detected belongs and the required false alarm probability; Data dimension reduction module: used to project the test data and auxiliary data into the signal subspace for data dimension reduction, so as to obtain new test data and new auxiliary data respectively; Parameter calculation module: used to calculate the sampling covariance matrix of the new auxiliary data using the new data to be measured, the new auxiliary data and the signal subspace matrix, calculate the power mismatch of the Gaussian colored noise in the data to be measured and the auxiliary data, calculate the coordinate vector of the target signal in the data to be measured in the subspace spanned by the signal subspace matrix, calculate the amplitude vector of the target signal in the data to be measured, and calculate the covariance matrix of the new Gaussian colored noise; Detection statistic calculation module: used to calculate the detection statistic based on the Wald criterion using the results output by the parameter calculation module; Detection threshold calculation module: used to calculate the detection threshold according to the false alarm probability; Target detection output module: used to compare the detection statistics and the detection threshold, and output the decision result of whether there is a target signal of interest in the data to be tested.

Images