Multi-radar cooperative target identification method based on adaptive manifold discrimination regression

Abstract

The invention provides a multi-radar cooperative target identification method based on adaptive manifold discrimination regression, and relates to the technical field of radar signal processing, and the method comprises the steps: determining a first data matrix according to the first HRRP data of at least two visual angles corresponding to a to-be-identified object; performing feature projection on the first data matrix according to the target projection matrix to obtain a first feature vector; the target projection matrix is obtained by utilizing second HRRP data of at least two view angles, training labels corresponding to the second HRRP data of each view angle and a first target function of multi-view angle identification to perform iterative updating on the initial projection matrix; target parameters of the first target function comprise view angle weights corresponding to all view angles, a projection matrix and a neighbor relation weight matrix; and determining an object category corresponding to the first HRRP data according to the first feature vector. According to the invention, the accuracy of multi-view HRRP target identification is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and in particular, to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression. Background Art

[0002] The fluctuation of high-resolution radar echoes reflects the distribution characteristics of the target along the radar line of sight, which is called the High Resolution Range Profile (HRRP). The target recognition method based on HRRP data occupies an important position in the field of radar target recognition. The existing traditional single-view HRRP recognition methods can be mainly divided into three methods: some methods extract aspect-invariant features, such as scattering center features, transform domain features, feature combination and optimization, and sparse representation features; some methods achieve framing in the angle domain to avoid excessive amplitude fluctuations within a frame; some methods use deep networks, such as Convolutional Neural Networks (CNNs), Recurrent Neural Network (RNN), and hybrid networks to extract angle-invariant features. In the scenario of multi-radar cooperative detection, the key problem of multi-view HRRP cooperative recognition is to model the data correlation and difference between multiple views.

[0003] Most of the current target recognition technologies for realizing multi-view HRRP joint processing are extensions of single-view HRRP methods. Some works use decision-level fusion methods or sparse representation to integrate multi-view HRRP data. For example, the target recognition of multi-view HRRP data is achieved by simply integrating the data of each view and performing weighted averaging, and the recognition accuracy is relatively low. Summary of the Invention

[0004] The present invention provides a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression to solve the defect of relatively low recognition accuracy in the prior art and realize the improvement of recognition accuracy.

[0005] In a first aspect, the present invention provides a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression, and the method includes the following steps: Obtain first HRRP data of at least two views corresponding to the object to be recognized; Determine a first data matrix according to the first HRRP data of each of the views; Perform eigen-projection on the first data matrix according to the target projection matrix to obtain the first eigenvector corresponding to the first data matrix; the target projection matrix is obtained by iteratively updating the initial projection matrix by using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function for multi-perspective recognition. The training labels are associated with the second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data. The target parameters of the first objective function include the perspective weights corresponding to each perspective, the projection matrix, and the neighbor relationship weight matrix. The neighbor relationship weight matrix is used to characterize the weights between neighboring samples in the preset categories under each perspective. Determine the object category corresponding to the first HRRP data according to the first eigenvector.

[0006] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, the target projection matrix is determined through the following steps, including: Construct the first objective function for multi-perspective recognition; Initialize the target parameters in the first objective function; When it is determined that the number of algorithm iterations has not reached the preset number of iterations, use the second HRRP data of at least two perspectives and the training labels corresponding to the second HRRP data of each perspective to iteratively update the target parameters to obtain the updated target parameters; Determine the updated projection matrix based on the updated target parameters; When it is determined that the number of algorithm iterations reaches the preset number of iterations, determine the updated projection matrix corresponding to the end of the iteration as the target projection matrix; the number of rows of the target projection matrix is determined based on the number of feature dimensions included in the second HRRP data, and the number of columns of the target projection matrix is determined based on the number of categories of preset objects.

[0007] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, the step of using the second HRRP data of at least two perspectives and the training labels corresponding to the second HRRP data of each perspective to iteratively update the target parameters to obtain the updated target parameters includes: Construct a second data matrix according to the second HRRP data of the at least two perspectives; For any one of the target parameters, take the other parameters except the target parameter as fixed parameters, update the first objective function according to the second data matrix, and solve to obtain each updated target parameter based on the updated second objective function.

[0008] A multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, wherein the target parameters further include an adjustment matrix, and the adjustment matrix is used for the adjustment term of adaptive weighted fusion; for any one of the target parameters, taking other parameters except the target parameter as fixed parameters, updating the first objective function according to the second data matrix, and based on the updated second objective function, solving to obtain each updated target parameter. According to the second data matrix, any one of the target parameters includes: Taking the neighbor relationship weight matrix, the adjustment matrix, and the perspective weights corresponding to each perspective in the first objective function as fixed parameters, updating the first objective function to obtain a second objective function; Calculating the gradient of the second objective function, and determining the projection matrix corresponding to when the gradient of the second objective function is 0 as the updated projection matrix; Taking the updated projection matrix, the adjustment matrix, and the perspective weights corresponding to each perspective as fixed parameters, updating the first objective function to obtain a third objective function; Calculating the gradient of the third objective function, and determining the neighbor relationship weight matrix corresponding to when the gradient of the third objective function is 0 as the updated neighbor relationship weight matrix; Taking the updated projection matrix, the updated neighbor relationship weight matrix, and the perspective weights corresponding to each perspective as fixed parameters, updating the first objective function to obtain a fourth objective function; Based on the fourth objective function, determining the updated adjustment matrix; Taking the updated projection matrix, the updated neighbor relationship weight matrix, and the updated adjustment matrix as fixed parameters, updating the first objective function to obtain a fifth objective function; Based on the fifth objective function, determining the updated perspective weights corresponding to each perspective; Taking the updated projection matrix, the updated neighbor relationship weight matrix, the updated adjustment matrix, and the updated perspective weights corresponding to each perspective as the updated target parameters.

[0009] A multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, wherein determining the first data matrix according to the first HRRP data of each perspective includes: Determining the first HRRP data of each feature dimension according to the first HRRP data of each perspective; Based on the first HRRP data of each feature dimension, determining the first data matrix.

[0010] A multi-radar collaborative target recognition method based on adaptive manifold discriminant regression according to the present invention, which projects the first data matrix according to the target projection matrix to obtain the first eigenvector corresponding to the first data matrix: Multiply the target projection matrix by the first data matrix to obtain an eigenvector matrix; Determine the first eigenvector based on the eigenvector matrix.

[0011] A multi-radar collaborative target recognition method based on adaptive manifold discriminant regression according to the present invention, determining the object category corresponding to the first HRRP data according to the first eigenvector, including: For any one of the first eigenvectors, determine the Euclidean distance between the first eigenvector and the second eigenvectors corresponding to the second HRRP data of each perspective; Based on the Euclidean distances between the first eigenvector and the second eigenvectors corresponding to the second HRRP data of each perspective, determine at least one neighbor closest to the first eigenvector; Based on the preset object categories corresponding to the at least one closest neighbor, determine the object category corresponding to the first eigenvector; Based on the object categories corresponding to each of the first eigenvectors, determine the object category corresponding to the first HRRP data.

[0012] In a second aspect, the present invention further provides a multi-radar collaborative target recognition device based on adaptive manifold discriminant regression, and the device includes the following modules: An acquisition module for acquiring the first HRRP data of at least two perspectives; An identification module for determining a first data matrix according to the first HRRP data of each perspective; Project the first data matrix according to the target projection matrix to obtain the first eigenvector corresponding to the first data matrix; the target projection matrix is obtained by iteratively updating the initial projection matrix using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function for multi-perspective recognition. The training labels are used to represent the second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors are used to determine the object categories corresponding to the second HRRP data; the target parameters of the first objective function include the perspective weights corresponding to each perspective, the projection matrix, and the neighbor relationship weight matrix, and the neighbor relationship weight matrix is used to represent the weights between neighbor samples in the preset categories under each perspective; Determine the object category corresponding to the first HRRP data according to the first eigenvector.

[0013] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for multi-radar collaborative target recognition based on adaptive manifold discriminant regression as described in any one of the above is implemented.

[0014] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for multi-radar collaborative target recognition based on adaptive manifold discriminant regression as described in any one of the above is implemented.

[0015] In a fifth aspect, the present invention further provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for multi-radar collaborative target recognition based on adaptive manifold discriminant regression as described in any one of the above is implemented.

[0016] The method for multi-radar collaborative target recognition based on adaptive manifold discriminant regression provided by the present invention obtains first HRRP data from at least two perspectives, and determines a first data matrix according to the first HRRP data of each perspective; then, performs feature projection on the first data matrix according to a target projection matrix to obtain a first eigenvector corresponding to the first data matrix, where the target projection matrix is obtained by iteratively updating an initial projection matrix using second HRRP data from at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function for multi-perspective recognition. The training labels are associated with second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data; furthermore, according to the first eigenvector, the object category corresponding to the first HRRP data is determined.

[0017] The target parameters of the first objective function in the present invention include an adjustment matrix, perspective weights corresponding to each of the perspectives, a projection matrix, and a neighbor relationship weight matrix. The neighbor relationship weight matrix is used to characterize the weights between neighbor samples in a preset category under each perspective. First, the initial projection matrix is iteratively updated using second HRRP data from at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function for multi-perspective recognition to obtain a target projection matrix. The training labels are used to characterize the object category corresponding to the second HRRP data of each perspective. In the present invention, the relationship between HRRP samples under different perspectives is fully considered, and the relationship between second HRRP data of different perspectives is modeled by adaptive weight reconstruction, thereby constructing the first objective function. Furthermore, feature projection is performed on the first HRRP data of multiple perspectives based on the target projection matrix to obtain a low-dimensional representation of the data, and finally, the accuracy of multi-perspective HRRP target recognition is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 FIG. is one of the schematic flowcharts of the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention.

[0020] Figure 2 FIG. is another schematic flowchart of the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention.

[0021] Figure 3 FIG. is yet another schematic flowchart of the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention.

[0022] Figure 4 FIG. is the schematic structural diagram of the multi-radar cooperative target recognition device based on adaptive manifold discriminant regression provided by the present invention.

[0023] Figure 5 FIG. is the schematic structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0025] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances so that the embodiments of this application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are usually of the same category, and the number of objects is not limited. For example, the first node can be one or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally means an "or" relationship between the associated objects before and after.

[0026] The following is combined withFigures 1 - 5 Describe the multi-radar collaborative target recognition method based on adaptive manifold discriminant regression of the present invention.

[0027] Figure 1 It is one of the flow schematic diagrams of the multi-radar collaborative target recognition method based on adaptive manifold discriminant regression provided by the present invention. As Figure 1 shown, this method includes the following: Step 101: Obtain the first HRRP data of at least two perspectives corresponding to the object to be recognized; Specifically, it should be noted first that the execution subject of this embodiment is an electronic device, which is used to realize the recognition of multi-perspective targets based on high-resolution range profile (HRRP) data of multiple perspectives, and improve the accuracy of target recognition.

[0028] First, obtain the first HRRP data of multiple perspectives corresponding to the object to be recognized. Among them, multiple perspectives refer to the perspectives corresponding to multi-radar collaborative detection. HRRP is the abbreviation of High Resolution Range Profile, which is a radar technology used to obtain the projection vector sum of the scattered sub-echoes of the target in the radar line-of-sight direction. HRRP contains many important structural information such as target size and scatter point distribution, which are crucial for radar target recognition and classification. In the field of radar target recognition, target recognition based on HRRP is an important research direction because it can provide the geometric structure information of the target and is a potential technology in radar automatic target recognition (RATR) applications. Generally speaking, target recognition based on HRRP is a key technology for target recognition and classification in the radar field. It provides the structural information of the target by analyzing the scatter point distribution of the target and is an important part of radar automatic target recognition technology.

[0029] The specific way to obtain the first HRRP data of multiple perspectives can be the HRRP obtained by using multiple radars with different perspectives to detect the detection target.

[0030] Step 102: Determine the first data matrix according to the first HRRP data of each perspective; Specifically, after obtaining the first HRRP data of each perspective, it is necessary to further perform target recognition based on the first HRRP data of multiple perspectives.

[0031] First, determine the first data matrix according to the first HRRP data of each perspective. The number of rows of the first data matrix is the feature dimension d after connecting all perspectives, and the number of columns is the total number of samples n, where: Among them, d k represents the feature dimension of the k-th perspective, d represents the feature dimension after connecting all perspectives, and v represents v perspectives, that is, v perspectives.

[0032] The first data matrix can be obtained as follows: Among them, represents the first HRRP data of the first perspective, represents the first HRRP data of the second perspective, represents the first HRRP data of the v-th perspective.

[0033] Step 103: Perform feature projection on the first data matrix according to the target projection matrix to obtain the first eigenvector corresponding to the first data matrix; the target projection matrix is obtained by iteratively updating the initial projection matrix using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function of multi-perspective recognition. The training labels are associated with the second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include the perspective weights corresponding to each perspective, the projection matrix, and the nearest neighbor relationship weight matrix, and the nearest neighbor relationship weight matrix is used to represent the weights between the nearest neighbor samples in the preset categories under each perspective; Specifically, among them, the training labels are used to represent the object categories corresponding to the second HRRP data of each perspective. Multiple object categories can be preset in advance. Different labels can correspond to different second eigenvectors. The training labels are associated with the second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data. The pre-constructed first objective function is, for example: Among them, λ 1 and λ 2 are trade-off parameters, represents the weighted projection matrix, S represents the nearest neighbor relationship weight matrix, represents the weight of the nearest neighbor between the m-th sample and the n-th sample in the category under the 1 -th perspective, α represents the perspective weights corresponding to each perspective, M = [M 2 , M n ∈ R n×cdenotes an adjustment matrix, such as a non - negative adjustment vector matrix, k represents the k - th perspective, and X k denotes the data sample matrix of HRRP for the k - th perspective, k denotes the projection transformation matrix of the weighted k - th perspective, Y denotes the sample label matrix, and ⊙ denotes the Hadamard product operation.

[0034] The manifold learning regularization term in the objective function can reveal the low - dimensional geometric structure of the data, which is used to capture the intrinsic nature of the data and explore the data distribution structure of multi - perspective HRRP. We use adaptive weights to measure the similarity of different neighbor pairs. In addition, HRRPs from different angles can be considered to have different data distributions and underlying patterns. Therefore, we introduce a manifold regularization term that models different perspectives separately to explore the geometric information of the data. Through manifold learning, features can be retained, and the projection of the geometric structure can be extracted, but this projection contains limited class information. To better serve the final target recognition task, a multi - perspective adaptive weighted fusion framework based on improved least - squares regression is proposed. After multi - perspective fusion, the obtained data features usually have redundant features. These redundant features are useless or even harmful to model training. To solve this problem, this method further imposes a row - sparsity constraint on the projection. By applying norm regularization to the projection matrix, we identify key feature groups and set the weights of other groups to zero.

[0035] The target projection matrix is obtained by iteratively updating the initial projection matrix using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function for multi - perspective recognition. For example, the target parameters are updated within a preset number of iterations, and the projection matrix corresponding to the end of the iteration is determined as the target projection matrix.

[0036] Furthermore, according to the target projection matrix, feature projection is performed on the first data matrix to obtain the first eigenvector corresponding to the first data matrix. For example, after obtaining the target projection matrix, the first eigenvector is extracted by multiplying the target projection matrix with the first data matrix.

[0037] Step 104: Determine the object category corresponding to the first HRRP data according to the first eigenvector.

[0038] Specifically, after obtaining the first eigenvector corresponding to the first data matrix, the prediction label corresponding to the first eigenvector, that is, the predicted object category corresponding to the target of the first HRRP data, can be determined to achieve high - precision target recognition.

[0039] For example, the KNN (K-Nearest Neighbors) algorithm is used on the first eigenvector to finally obtain the predicted label (predicted object category) of the HRRP sample.

[0040] The method provided in this embodiment obtains first HRRP data from at least two perspectives, and determines a first data matrix based on the first HRRP data from each perspective; then, based on the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix, wherein the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using the second HRRP data from at least two perspectives, the training labels corresponding to the second HRRP data from each perspective, and the first objective function of multi-perspective recognition, the training labels are associated with the second eigenvectors corresponding to the second HRRP data from each perspective, and the second eigenvectors corresponding to the second HRRP data from each perspective are used to determine the object category corresponding to the second HRRP data; and then, based on the first eigenvector, the object category corresponding to the first HRRP data is determined.

[0041] The target parameters of the first objective function in the present invention include an adjustment matrix, perspective weights corresponding to each of the perspectives, a projection matrix, and a neighbor relationship weight matrix. The neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in preset categories under each perspective. First, the initial projection matrix is ​​iteratively updated using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function of multi-perspective recognition to obtain a target projection matrix. The training labels are used to characterize the object categories corresponding to the second HRRP data of each perspective. The present invention fully considers the relationship between HRRP samples under different perspectives, and constructs the first objective function by adaptively modeling the relationship between the second HRRP data of different perspectives. Furthermore, based on the target projection matrix, the first HRRP data of multiple perspectives are feature projected to obtain a low-dimensional representation of the data, thereby ultimately improving the accuracy of multi-perspective HRRP target recognition.

[0042] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, the target projection matrix is ​​determined by the following steps, including: Construct the first objective function of multi-view recognition; Initialize the target parameters in the first target function; When it is determined that the number of algorithm iterations has not reached the preset number of iterations, iteratively updating the target parameters using the second HRRP data of at least two perspectives and the training labels corresponding to the second HRRP data of each perspective to obtain updated target parameters; Based on the updated target parameters, determining an updated projection matrix; When it is determined that the number of algorithm iterations reaches the preset number of iterations, the updated projection matrix corresponding to the termination of the iteration is determined as the target projection matrix; the number of rows of the target projection matrix is ​​determined based on the number of perspectives contained in the second HRRP data, and the number of columns of the target projection matrix is ​​determined based on the number of preset object categories.

[0043] Specifically, in some embodiments, the target projection matrix is ​​determined by the following steps, including: First, the first objective function of multi-view recognition is constructed. The first objective function is exemplified as follows: Among them, λ 1 and λ 2 is a trade-off parameter, represents the weighted projection matrix, S represents the neighbor relationship weight matrix, Indicates From a perspective The weight of the nearest neighbor between the m-th sample and the n-th sample in the class, α represents the perspective weight corresponding to each perspective, M=[ M 1 , M 2 , …M n ]∈R n×c represents an adjustment matrix, such as a non-negative adjustment vector matrix, k represents the kth viewing angle, X k Represents the data sample matrix of HRRP of the k-th perspective, k represents the weighted projection transformation matrix of the kth perspective, and Y represents the sample label matrix.

[0044] Among them, the first objective function includes a manifold regularization term cost function based on adaptive neighbor weights and a multi-view adaptive weighted fusion cost function. The manifold regularization term cost function evaluates the similarity of different neighbor pairs by introducing adaptive weights, and models different perspectives separately to learn the distribution patterns of HRRP at different angles. The multi-view adaptive weighted fusion cost function further fuses the obtained manifold projection based on the improved least squares regression method, and introduces supervisory information to better serve the recognition task. The relationship between different perspectives is modeled through adaptive weights, so as to more fully model the consistency, complementarity and difference of different perspectives. It can effectively improve the recognition performance.

[0045] Furthermore, the multi-view high-resolution range profile HRRP is input to initialize the target parameters in the first objective function. The parameters to be initialized are: adjustment matrix M, view weight α k And the projection matrix .

[0046] First, the adjustment matrix is ​​initialized to M=0, and the projection matrix is ​​initialized to 0 =0, initialize the view weight to In addition, for the manifold regularization term, we first use the Laplacian matrix of the data sample to initialize the neighbor relationship weight matrix and construct it as a k-nearest neighbor graph with binary values, that is, .

[0047] Further, when it is determined that the number of algorithm iterations has not reached the preset number of iterations, the target parameters are iteratively updated using the second HRRP data of at least two perspectives and the training labels corresponding to the second HRRP data of each perspective to obtain updated target parameters. The preset number of iterations can be set based on the scale of the data sample, for example, 100 times, 500 times, 1000 times, until the function converges. Based on the updated target parameters, an updated projection matrix is ​​determined; Further, when it is determined that the number of algorithm iterations reaches a preset number of iterations, the updated projection matrix corresponding to the termination of the iteration is determined as the target projection matrix, wherein the number of rows of the target projection matrix is ​​determined based on the number of feature dimensions included in the second HRRP data, and the number of columns of the target projection matrix is ​​determined based on the number of preset object categories.

[0048] For example, Figure 2 FIG. 2 is a flow chart of the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, which shows the process of solving the algorithm to obtain the target projection matrix, such as Figure 2 As shown, the method includes: First, input multi-view HRRP; Initialize adjustment matrix, weight matrix, and projection matrix; Determine whether the number of iterations exceeds the limit; If the number of iterations is within the limit, update the parameters; determine whether the difference meets the standard; if it does, stop the iteration and output the projection matrix; if it does not, add 1 to the number of iterations and re-determine whether the number of iterations exceeds the limit; If the number of iterations exceeds the limit, the iteration is stopped.

[0049] The method provided in this embodiment first constructs a first objective function for multi-view recognition, where the first objective function includes a manifold regularization term cost function based on adaptive neighbor weights and a multi-view adaptive weighted fusion cost function, initializes the target parameters in the first objective function, and iteratively updates the algorithm using the second HRRP data of at least two viewpoints and the training labels corresponding to the second HRRP data of each viewpoint to obtain updated target parameters; when it is determined that the number of algorithm iterations reaches a preset number of iterations, the updated projection matrix corresponding to the termination of the iteration is determined as the target projection matrix. The present invention learns a target projection matrix through adaptive manifold learning, adaptive weighted fusion and L2,1 norm regularization. The target projection matrix can be used to obtain a low-dimensional representation of multi-view HRRP data, ultimately achieving multi-view HRRP target recognition and improving the accuracy of target recognition.

[0050] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, second HRRP data of at least two perspectives and training labels corresponding to the second HRRP data of each perspective are used to iteratively update target parameters to obtain updated target parameters, including: constructing a second data matrix according to the second HRRP data of at least two perspectives; For any target parameter, other parameters except the target parameter are taken as fixed parameters, the first target function is updated according to the second data matrix, and each updated target parameter is solved based on the updated second target function.

[0051] Specifically, in some embodiments, obtaining the second HRRP data using at least two viewing angles in the target projection matrix and the training labels corresponding to the second HRRP data of each viewing angle, and iteratively updating the target parameters to obtain the updated target parameters includes the following steps: First, construct a second data matrix according to the second HRRP data of at least two viewing angles; Furthermore, for any target parameter, other parameters other than the target parameter are used as fixed parameters, the first target function is updated according to the second data matrix, and each updated target parameter is obtained based on the updated second target function. The target parameters include the adjustment matrix M, the perspective weights corresponding to each perspective , projection matrix And a neighbor relationship weight matrix S, where the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in the preset categories under each perspective.

[0052] For example, the neighbor relationship weight matrix S, the perspective weight corresponding to each perspective And update the first objective function with the adjustment matrix M as a fixed parameter to obtain the second objective function. Furthermore, by calculating the gradient of the second objective function and setting the gradient to 0, the projection matrix corresponding to the gradient of the second objective function being 0 is obtained. It is determined as the updated target parameter. The update processes of other target parameters are similar.

[0053] The method provided in this embodiment first constructs a second data matrix based on the second HRRP data of at least two perspectives. Then, for any target parameter, other parameters except the target parameter are used as fixed parameters, and the first objective function is updated according to the second data matrix. Based on the updated second objective function, each updated target parameter is solved, realizing the iterative update of the parameters. The updated target parameter corresponding to the preset number of iterations of the algorithm is determined as the final parameter value, such as the target projection matrix. Finally, eigenvector extraction is performed based on the target projection matrix, and target recognition is finally realized.

[0054] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, the target parameter further includes an adjustment matrix, and the adjustment matrix is used for the adjustment term of adaptive weighted fusion; for any target parameter, other parameters except the target parameter are used as fixed parameters, and the first objective function is updated according to the second data matrix. Based on the updated second objective function, each updated target parameter is solved. According to the second data matrix, any parameter in the target parameter includes: Taking the neighbor relationship weight matrix, adjustment matrix, and perspective weights corresponding to each perspective in the first objective function as fixed parameters, the first objective function is updated to obtain the second objective function; Calculate the gradient of the second objective function, and determine the projection matrix corresponding to the gradient of the second objective function being 0 as the updated projection matrix; Taking the updated projection matrix, adjustment matrix, and perspective weights corresponding to each perspective as fixed parameters, the first objective function is updated to obtain the third objective function; Calculate the gradient of the third objective function, and determine the neighbor relationship weight matrix corresponding to the gradient of the third objective function being 0 as the updated neighbor relationship weight matrix; Taking the updated projection matrix, updated neighbor relationship weight matrix, and perspective weights corresponding to each perspective as fixed parameters, the first objective function is updated to obtain the fourth objective function; Based on the fourth objective function, the updated adjustment matrix is determined; Taking the updated projection matrix, updated neighbor relationship weight matrix, and updated adjustment matrix as fixed parameters, the first objective function is updated to obtain the fifth objective function; Based on the fifth objective function, the updated perspective weights corresponding to each perspective are determined; The updated projection matrix, the updated neighbor relationship weight matrix, the updated adjustment matrix, and the updated viewing angle weights corresponding to each viewing angle are determined as updated target parameters.

[0055] Specifically, in some embodiments, for any target parameter, other parameters other than the target parameter are used as fixed parameters, the first target function is updated according to the second data matrix, and each updated target parameter is obtained based on the updated second target function. The specific implementation process of any parameter in the target parameter according to the second data matrix includes the following steps: First, the neighbor relationship weight matrix, the adjustment matrix, and the perspective weights corresponding to each perspective in the first objective function are used as fixed parameters, the first objective function is updated to obtain the second objective function, the gradient of the second objective function is calculated, and the projection matrix corresponding to the gradient of the second objective function being 0 is determined as the updated projection matrix; The second objective function is, for example: in, is a vector. By calculating the gradient of the second objective function and making the gradient of the second objective function equal to 0, we can obtain the value that minimizes the second objective function. , which is expressed as follows: in, is a diagonal matrix whose diagonal elements are , , where T is as follows: Further, the updated projection matrix, the adjustment matrix, and the perspective weights corresponding to each perspective are used as fixed parameters to update the first objective function to obtain a third objective function; the gradient of the third objective function is calculated, and the neighbor relationship weight matrix corresponding to the gradient of the third objective function is 0, which is determined as the updated neighbor relationship weight matrix, wherein the third objective function is, for example: The solution to the optimization problem can be obtained as: Furthermore, the updated projection matrix M and the updated neighbor relationship weight matrix , the perspective weight corresponding to each perspective As a fixed parameter, the first objective function is updated to obtain a fourth objective function, and then, based on the fourth objective function, an updated adjustment matrix is ​​determined. For example, the fourth objective function is: Let the residual be , and based on the Frobenius norm theory, the problem can be decoupled into n×c sub-problems. For the elements in the i-th row and j-th column of matrices , and M , , . The problem can be reformulated as: s.t. Based on being composed only of the elements of " ", we can obtain: s.t. It is easy to obtain that the solution of the adjustment matrix is: Furthermore, taking the updated projection matrix, the updated neighbor relationship weight matrix, and the updated adjustment matrix as fixed parameters, the first objective function is updated to obtain the fifth objective function; based on the fifth objective function, the perspective weights corresponding to the updated perspectives are determined; For example, with the fixed parameters , the fifth objective function to be optimized is: s.t. , Based on the Cauchy inequality theory, the solution can be obtained as: Furthermore, the updated projection matrix, the updated neighbor relationship weight matrix, the updated adjustment matrix, and the perspective weights corresponding to the updated perspectives are determined as the updated target parameters, realizing the update of the target parameters. Finally, based on the updated target parameters corresponding to the satisfaction of the iteration termination, the target projection matrix is determined, and the first data matrix is eigen-projected based on the target projection matrix, and target recognition is performed using the projected eigenvectors.

[0056] In the method provided in this embodiment, the neighbor relationship weight matrix, adjustment matrix, and view weights corresponding to each view in the first objective function are used as fixed parameters to update the first objective function, obtaining a second objective function; calculating the gradient of the second objective function, and determining the projection matrix corresponding to when the gradient of the second objective function is 0 as the updated projection matrix; using the updated projection matrix, adjustment matrix, and view weights corresponding to each view as fixed parameters to update the first objective function, obtaining a third objective function; calculating the gradient of the third objective function, and determining the neighbor relationship weight matrix corresponding to when the gradient of the third objective function is 0 as the updated neighbor relationship weight matrix; using the updated projection matrix, updated neighbor relationship weight matrix, and view weights corresponding to each view as fixed parameters to update the first objective function, obtaining a fourth objective function; determining the updated adjustment matrix based on the fourth objective function; using the updated projection matrix, updated neighbor relationship weight matrix, and updated adjustment matrix as fixed parameters to update the first objective function, obtaining a fifth objective function; determining the updated view weights corresponding to each view based on the fifth objective function; determining the updated projection matrix, updated neighbor relationship weight matrix, updated adjustment matrix, and updated view weights corresponding to each view as the updated target parameters. Finally, based on the updated target parameters corresponding to when the iteration terminates, determining the target projection matrix, performing feature projection on the first data matrix based on the target projection matrix, and using the projected feature vectors for target recognition, improving the accuracy of multi-view HRRP target recognition.

[0057] According to a multi-radar collaborative target recognition method based on adaptive manifold discriminant regression provided by the present invention, determining a first data matrix according to the first HRRP data of each view includes: Determining the first HRRP data of each feature dimension according to the first HRRP data of each view; Based on the first HRRP data of each feature dimension, determining the first data matrix.

[0058] Specifically, in some embodiments, the specific implementation process of determining the first data matrix according to the first HRRP data of each view in step 102 includes the following steps: First, determine the first HRRP data of each feature dimension according to the first HRRP data of each view. Among them, the feature dimension is the feature dimension d after connecting all views, where: where d k represents the feature dimension of the kth view, d represents the feature dimension after connecting all views, and v represents v views, that is, v views.

[0059] Furthermore, based on the first HRRP data of each feature dimension, a first data matrix is ​​determined. The first HRRP data of each feature dimension is placed in a matrix, where the rows of the matrix represent the feature dimensions and the columns of the matrix represent the number of samples.

[0060] The method provided in this embodiment determines the first data matrix based on the first HRRP data of each perspective, and then performs feature projection on the first data matrix through the target projection matrix, and finally realizes multi-perspective HRRP target recognition based on the projected feature vector, thereby improving the accuracy of target recognition.

[0061] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, feature projection is performed on a first data matrix according to a target projection matrix to obtain a first feature vector corresponding to the first data matrix: Multiply the target projection matrix by the first data matrix to obtain an eigenvector matrix; Based on the eigenvector matrix, a first eigenvector is determined.

[0062] Specifically, in some embodiments, step 103 may be implemented by the following steps: Among them, the target projection matrix is, for example: The first data matrix is, for example: The target projection matrix is ​​multiplied by the first data matrix to obtain an eigenvector matrix, wherein the eigenvector matrix is ​​an n×c matrix. Further, based on the eigenvector matrix, the first eigenvector can be extracted.

[0063] The method provided in this embodiment performs feature projection on the first data matrix according to the target projection matrix to obtain a first eigenvector corresponding to the first data matrix. Furthermore, the object category corresponding to the first HRRP data can be predicted based on the first eigenvector, thereby improving the accuracy of object recognition.

[0064] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, determining the object category corresponding to the first HRRP data according to the first feature vector includes: For any first eigenvector, determining the Euclidean distance between the first eigenvector and the second eigenvector corresponding to the second HRRP data of each viewing angle; Determine at least one neighbor closest to the first eigenvector based on the Euclidean distance between the first eigenvector and the second eigenvector corresponding to the second HRRP data of each viewing angle; Determining the object category corresponding to the first feature vector based on a preset object category corresponding to at least one nearest neighbor; Based on the object categories corresponding to the first feature vectors, the object category corresponding to the first HRRP data is determined.

[0065] Specifically, the specific implementation process of determining the object category corresponding to the first HRRP data according to the first feature vector in step 104 may include the following steps: For example, the KNN algorithm is used to predict the predicted label of HRRP data, that is, the object category. The K nearest neighbor algorithm is a basic and widely used classification and regression method. It makes predictions by measuring the distance between different feature values.

[0066] First, it should be noted that there may be multiple first eigenvectors, and for any eigenvector, the Euclidean distance between the first eigenvector and the second eigenvector corresponding to the second HRRP data of each viewing angle is determined.

[0067] Furthermore, based on the Euclidean distance between the feature vector and the second feature vector corresponding to the second HRRP data of each perspective, at least one neighbor closest to the feature vector can be determined, wherein the number of nearest neighbors is at least 1, can be 3, 5, and generally an odd number of neighbors is used.

[0068] Further, the object category corresponding to the first feature vector is determined based on the preset object category corresponding to at least one of the nearest neighbors. For example, the preset object category corresponding to the neighbor with the smallest distance value is determined as the object category corresponding to the first feature vector.

[0069] Further, based on the object categories corresponding to the first feature vectors, the object category corresponding to the first HRRP data is determined. The KNN algorithm classifies the test data point into the category by checking which category is the most common among the K nearest neighbors.

[0070] The method provided in this embodiment uses the K nearest neighbor algorithm on the feature vector, and finally obtains the predicted label of the multi-view HRRP data, that is, performs target recognition, thereby improving the accuracy of target recognition of the multi-view data.

[0071] Figure 3 FIG. 3 is a flow chart of a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention. Figure 3 As shown, the method includes: Enter HRRP; Locality-preserving manifold learning; L2,1 norm (row sparsity) constraint; Feature projection; The nearest neighbor algorithm outputs a predicted label.

[0072] Firstly, the current multi-view HRRP recognition method cannot effectively explore the relationship and unique distribution pattern of data samples from different viewpoints. Simple multi-view fusion cannot effectively eliminate posture sensitivity and may lead to reduced accuracy. The present invention proposes a manifold regularized cost function based on adaptive neighbor weights, which introduces adaptive weights to evaluate the similarity of different neighbor pairs, and models different viewpoints separately to achieve learning of the distribution patterns of HRRP at different angles.

[0073] In addition, the existing multi-view HRRP recognition methods focus on the fusion of HRRP signals from multiple viewpoints, without considering the relationship between HRRP samples at different viewpoints, and without accurately and effectively modeling the relationship between data from different viewpoints. The present invention proposes a multi-view adaptive weighted fusion cost function. Based on the improved least squares regression method, supervision information is introduced to further fuse the obtained manifold projection. The relationship between different viewpoints is modeled by adaptive weights, so as to more fully model the consistency, complementarity and difference of different viewpoints. At the same time, a norm regularization constraint is imposed on the projection matrix to remove redundant information and further improve the recognition performance.

[0074] The multi-radar collaborative target recognition device based on adaptive manifold discriminant regression provided by the present invention is described below. The multi-radar collaborative target recognition device based on adaptive manifold discriminant regression described below and the multi-radar collaborative target recognition method based on adaptive manifold discriminant regression described above can be referenced to each other.

[0075] Figure 4 : is a schematic diagram of the structure of a multi-radar cooperative target recognition device based on adaptive manifold discriminant regression provided by the present invention, such as Figure 4 As shown, the multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression includes: An acquisition module 410 is used to acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; An identification module 420, configured to determine a first data matrix according to the first HRRP data of each viewing angle; According to the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix; the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using second HRRP data of at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function of multi-perspective recognition, the training labels are used to characterize the second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include perspective weights corresponding to each perspective, a projection matrix, and a neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in a preset category under each perspective; An object category corresponding to the first HRRP data is determined according to the first feature vector.

[0076] The device provided in this embodiment includes an acquisition module 410 and an identification module 420, the acquisition module 410 is used to acquire first HRRP data of at least two perspectives, and the identification module 420 is used to determine a first data matrix based on the first HRRP data of each perspective; then, according to the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix, wherein the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using second HRRP data of at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function of multi-perspective recognition, and the training labels are used to characterize the object category corresponding to the second HRRP data of each perspective; and then, according to the first eigenvector, the object category corresponding to the first HRRP data is determined.

[0077] The target parameters of the first objective function in the present invention include an adjustment matrix, perspective weights corresponding to each of the perspectives, a projection matrix, and a neighbor relationship weight matrix. The neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in preset categories under each perspective. First, the initial projection matrix is ​​iteratively updated using the second HRRP data of at least two perspectives, the training labels corresponding to the second HRRP data of each perspective, and the first objective function of multi-perspective recognition to obtain a target projection matrix. The training labels are used to characterize the object categories corresponding to the second HRRP data of each perspective. The present invention fully considers the relationship between HRRP samples under different perspectives, and constructs the first objective function by adaptively modeling the relationship between the second HRRP data of different perspectives. Furthermore, based on the target projection matrix, the first HRRP data of multiple perspectives are feature projected to obtain a low-dimensional representation of the data, thereby ultimately improving the accuracy of multi-perspective HRRP target recognition.

[0078] According to a multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression provided by the present invention, the device also includes a projection module; the projection module is specifically used to: The target projection matrix is ​​determined by the following steps, including: Constructing a first objective function of the multi-view recognition; Initializing target parameters in the first target function; When it is determined that the number of algorithm iterations has not reached the preset number of iterations, iteratively updating the target parameters using the second HRRP data of at least two viewing angles and the training labels corresponding to the second HRRP data of each viewing angle to obtain updated target parameters; Based on the updated target parameters, determining an updated projection matrix; When it is determined that the number of algorithm iterations reaches the preset number of iterations, the updated projection matrix corresponding to the termination of the iteration is determined as the target projection matrix; the number of rows of the target projection matrix is ​​determined based on the number of viewing angles contained in the second HRRP data, and the number of columns of the target projection matrix is ​​determined based on the number of preset object categories.

[0079] According to a multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression provided by the present invention, the projection module is further used for: constructing a second data matrix according to the second HRRP data of the at least two viewing angles; For any of the target parameters, other parameters except the target parameter are taken as fixed parameters, the first target function is updated according to the second data matrix, and each updated target parameter is solved based on the updated second target function.

[0080] According to a multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression provided by the present invention, the target parameters also include an adjustment matrix, and the adjustment matrix is ​​used for the adjustment item of adaptive weighted fusion; the projection module is also used for: Taking the neighbor relationship weight matrix, the adjustment matrix, and the perspective weights corresponding to each perspective in the first objective function as fixed parameters, the first objective function is updated to obtain a second objective function; Calculating the gradient of the second objective function, and determining the projection matrix corresponding to when the gradient of the second objective function is 0 as the updated projection matrix; Using the updated projection matrix, the adjustment matrix, and the perspective weights corresponding to the perspectives as fixed parameters, the first objective function is updated to obtain a third objective function; Calculating the gradient of the third objective function, and determining the neighbor relationship weight matrix corresponding to when the gradient of the third objective function is 0 as the updated neighbor relationship weight matrix; Using the updated projection matrix, the updated neighbor relationship weight matrix, and the perspective weights corresponding to the perspectives as fixed parameters, the first objective function is updated to obtain a fourth objective function; Based on the fourth objective function, determining an updated adjustment matrix; Using the updated projection matrix, the updated neighbor relationship weight matrix, and the updated adjustment matrix as fixed parameters, the first objective function is updated to obtain a fifth objective function; Based on the fifth objective function, determining updated perspective weights corresponding to each of the perspectives; The updated projection matrix, the updated neighbor relationship weight matrix, the updated adjustment matrix, and the updated viewing angle weights corresponding to each of the viewing angles are determined as the updated target parameters.

[0081] According to a multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression provided by the present invention, the recognition module 420 is specifically used for: Determining first HRRP data of each feature dimension according to the first HRRP data of each viewing angle; The first data matrix is ​​determined based on the first HRRP data of each of the characteristic dimensions.

[0082] According to a multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the present invention, the first data matrix is ​​feature projected according to the target projection matrix to obtain a first feature vector corresponding to the first data matrix: Multiplying the target projection matrix by the first data matrix to obtain an eigenvector matrix; Based on the eigenvector matrix, the first eigenvector is determined.

[0083] According to the multi-radar cooperative target recognition device 400 based on adaptive manifold discriminant regression provided by the present invention, the recognition module 420 is further used for: For any of the first eigenvectors, determining a Euclidean distance between the first eigenvector and a second eigenvector corresponding to the second HRRP data of each of the viewing angles; determining at least one neighbor closest to the first eigenvector based on a Euclidean distance between the first eigenvector and a second eigenvector corresponding to the second HRRP data of each of the viewing angles; Determining the object category corresponding to the first feature vector based on the preset object category corresponding to the at least one nearest neighbor; Based on the object categories corresponding to the first feature vectors, the object category corresponding to the first HRRP data is determined.

[0084] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression, and the method includes: Acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; Determining a first data matrix according to the first HRRP data of each of the viewing angles; According to the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix; the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using second HRRP data of at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function of multi-perspective recognition, the training labels are associated with second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include perspective weights corresponding to each perspective, a projection matrix, and a neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in a preset category under each perspective; An object category corresponding to the first HRRP data is determined according to the first feature vector.

[0085] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0086] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the above methods, the method includes: Acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; Determining a first data matrix according to the first HRRP data of each of the viewing angles; According to the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix; the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using second HRRP data of at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function of multi-perspective recognition, the training labels are associated with second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include perspective weights corresponding to each perspective, a projection matrix, and a neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in a preset category under each perspective; An object category corresponding to the first HRRP data is determined according to the first feature vector.

[0087] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression provided by the above methods, the method comprising: Acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; Determining a first data matrix according to the first HRRP data of each of the viewing angles; According to the target projection matrix, feature projection is performed on the first data matrix to obtain a first eigenvector corresponding to the first data matrix; the target projection matrix is ​​obtained by iteratively updating the initial projection matrix using second HRRP data of at least two perspectives, training labels corresponding to the second HRRP data of each perspective, and a first objective function of multi-perspective recognition, the training labels are associated with second eigenvectors corresponding to the second HRRP data of each perspective, and the second eigenvectors corresponding to the second HRRP data of each perspective are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include perspective weights corresponding to each perspective, a projection matrix, and a neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in a preset category under each perspective; An object category corresponding to the first HRRP data is determined according to the first feature vector.

[0088] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0089] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-radar cooperative target recognition method based on adaptive manifold discriminant regression, characterized in that: include: Acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; Determining a first data matrix according to the first HRRP data of each of the viewing angles; Performing feature projection on the first data matrix according to the target projection matrix to obtain a first feature vector corresponding to the first data matrix; The target projection matrix is ​​obtained by iteratively updating the initial projection matrix using the second HRRP data of at least two viewing angles, the training labels corresponding to the second HRRP data of each viewing angle, and the first objective function of multi-view recognition, wherein the training labels are associated with the second eigenvectors corresponding to the second HRRP data of each viewing angle, and the second eigenvectors corresponding to the second HRRP data of each viewing angle are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include the viewing angle weights corresponding to each viewing angle, the projection matrix, and the neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in the preset category under each viewing angle; An object category corresponding to the first HRRP data is determined according to the first feature vector.

2. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 1 is characterized in that: The target projection matrix is ​​determined by the following steps, including: Constructing a first objective function of the multi-view recognition; Initializing target parameters in the first target function; When it is determined that the number of algorithm iterations has not reached the preset number of iterations, iteratively updating the target parameters using the second HRRP data of at least two viewing angles and the training labels corresponding to the second HRRP data of each viewing angle to obtain updated target parameters; Based on the updated target parameters, determining an updated projection matrix; When it is determined that the number of algorithm iterations reaches the preset number of iterations, the updated projection matrix corresponding to the termination of the iteration is determined as the target projection matrix; the number of rows of the target projection matrix is ​​determined based on the number of feature dimensions contained in the second HRRP data, and the number of columns of the target projection matrix is ​​determined based on the number of preset object categories.

3. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 1 is characterized in that: The step of iteratively updating the target parameter by using the second HRRP data of at least two viewing angles and the training labels corresponding to the second HRRP data of each viewing angle to obtain the updated target parameter includes: constructing a second data matrix according to the second HRRP data of the at least two viewing angles; For any of the target parameters, other parameters except the target parameter are taken as fixed parameters, the first target function is updated according to the second data matrix, and each updated target parameter is solved based on the updated second target function.

4. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 3 is characterized in that: The target parameters also include an adjustment matrix, which is used for adjustment items of adaptive weighted fusion; for any of the target parameters, other parameters other than the target parameters are used as fixed parameters, the first objective function is updated according to the second data matrix, and each updated target parameter is obtained by solving the updated second objective function based on the updated second data matrix, and any parameter in the target parameters is adjusted according to the second data matrix, including: Taking the neighbor relationship weight matrix, the adjustment matrix, and the perspective weights corresponding to each perspective in the first objective function as fixed parameters, the first objective function is updated to obtain a second objective function; Calculating the gradient of the second objective function, and determining the projection matrix corresponding to when the gradient of the second objective function is 0 as the updated projection matrix; Using the updated projection matrix, the adjustment matrix, and the perspective weights corresponding to the perspectives as fixed parameters, the first objective function is updated to obtain a third objective function; Calculating the gradient of the third objective function, and determining the neighbor relationship weight matrix corresponding to when the gradient of the third objective function is 0 as the updated neighbor relationship weight matrix; Using the updated projection matrix, the updated neighbor relationship weight matrix, and the perspective weights corresponding to the perspectives as fixed parameters, the first objective function is updated to obtain a fourth objective function; Based on the fourth objective function, determining an updated adjustment matrix; Using the updated projection matrix, the updated neighbor relationship weight matrix, and the updated adjustment matrix as fixed parameters, the first objective function is updated to obtain a fifth objective function; Based on the fifth objective function, determining updated perspective weights corresponding to each of the perspectives; The updated projection matrix, the updated neighbor relationship weight matrix, the updated adjustment matrix, and the updated viewing angle weights corresponding to each of the viewing angles are determined as the updated target parameters.

5. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 1 is characterized in that: The determining of the first data matrix according to the first HRRP data of each viewing angle includes: Determining first HRRP data of each feature dimension according to the first HRRP data of each viewing angle; The first data matrix is ​​determined based on the first HRRP data of each of the characteristic dimensions.

6. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 1 is characterized in that: According to the target projection matrix, feature projection is performed on the first data matrix to obtain a first feature vector corresponding to the first data matrix: Multiplying the target projection matrix by the first data matrix to obtain an eigenvector matrix; Based on the eigenvector matrix, the first eigenvector is determined.

7. The multi-radar cooperative target recognition method based on adaptive manifold discriminant regression according to claim 1 is characterized in that: Determining, according to the first feature vector, an object category corresponding to the first HRRP data, includes: For any of the first eigenvectors, determining a Euclidean distance between the first eigenvector and a second eigenvector corresponding to the second HRRP data of each of the viewing angles; determining at least one neighbor closest to the first eigenvector based on a Euclidean distance between the first eigenvector and a second eigenvector corresponding to the second HRRP data of each of the viewing angles; Determining the object category corresponding to the first feature vector based on the preset object category corresponding to the at least one nearest neighbor; Based on the object categories corresponding to the first feature vectors, the object category corresponding to the first HRRP data is determined.

8. A multi-radar cooperative target recognition device based on adaptive manifold discriminant regression, characterized in that: include: An acquisition module, used to acquire first HRRP data of at least two viewing angles corresponding to the object to be identified; an identification module, configured to determine a first data matrix according to the first HRRP data of each of the viewing angles; Performing feature projection on the first data matrix according to the target projection matrix to obtain a first feature vector corresponding to the first data matrix; The target projection matrix is ​​obtained by iteratively updating the initial projection matrix using the second HRRP data of at least two viewing angles, the training labels corresponding to the second HRRP data of each viewing angle, and the first objective function of multi-view recognition, wherein the training labels are associated with the second eigenvectors corresponding to the second HRRP data of each viewing angle, and the second eigenvectors corresponding to the second HRRP data of each viewing angle are used to determine the object category corresponding to the second HRRP data; the target parameters of the first objective function include the viewing angle weights corresponding to each viewing angle, the projection matrix, and the neighbor relationship weight matrix, and the neighbor relationship weight matrix is ​​used to characterize the weights between neighbor samples in the preset category under each viewing angle; An object category corresponding to the first HRRP data is determined according to the first feature vector.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-radar collaborative target recognition method based on adaptive manifold discriminant regression as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-radar cooperative target recognition method based on adaptive manifold discriminant regression as claimed in any one of claims 1 to 7 is implemented.

Images