Small sample radar HRRP target identification method based on optimal transmission measurement

Abstract

The invention discloses a small sample radar HRRP target identification method based on optimal transmission measurement, and the method comprises the steps: obtaining an HRRP signal of a to-be-detected target, and obtaining a query set HRPP feature; based on the optimal transmission scheme, carrying out iterative solution on the transmission matrix of the to-be-detected target to obtain an optimal transmission matrix; wherein the transmission matrix is used for representing the probability that the HRPP features are classified into various categories, the HRRP features comprise query set HRPP features and support set HRRP features with known category labels, the category center of each category is updated along with iteration of the transmission matrix, and the category center is a typical feature value of the category to which the category center belongs; and determining the category of the to-be-detected target according to the optimal transmission matrix. According to the method, the inherent attitude sensitivity influence of the HRRP data can be overcome, and under the condition of extremely small samples, robust recognition is carried out based on the obtained HRRP signals of different attitudes of the same target.

Description

Technical Field

[0001] This invention belongs to the field of target recognition technology, specifically relating to a small-sample radar HRRP target recognition method based on optimal transmission metric. Background Technology

[0002] Radar target recognition is a crucial technology playing a vital role in fields such as intelligent driving and aerospace. High-resolution range profiles (HRRPs) have become a key technology in radar target recognition due to their ease of acquisition and rich content of target structure information. However, in practical applications, especially in aircraft recognition where targets are constantly maneuvering, the amount of radar data is scarce, and severe class imbalance exists, the inherent attitude sensitivity of HRRP data and the limited number of training samples have become bottlenecks restricting the development of target recognition technology. Specifically, the inherent attitude sensitivity of HRRP data refers to the characteristic that the shape of the acquired HRRP changes drastically when the observation angle (i.e., attitude angle, usually azimuth angle) of the same target relative to the radar changes.

[0003] To address this issue, researchers typically employ methods such as data augmentation, few-shot learning, and transfer learning to effectively utilize scarce high-resolution distance image data. While traditional techniques attempt to improve target recognition performance based on HRRP from different angles, none have fundamentally solved the fundamental problem arising from the intertwined challenges of "the inherent pose sensitivity of HRRP data" and "the scarcity of training samples."

[0004] Specifically, data augmentation methods based on complex deep models typically require a large amount of labeled training data to train the model, making them unsuitable for situations where the number of training samples is sparse. Furthermore, this method requires the input data to be a continuous HRRP sequence. When the target moves rapidly or is observed by radar, the HRRP sequence may become incomplete or unstable, affecting the model's recognition performance.

[0005] The few-shot learning method based on supervised contrastive learning constructs positive and negative sample pairs using known azimuth information and introduces a new loss function to train the model. While this method can reduce the need for labeled samples to some extent, its ability to construct positive sample pairs is limited when the azimuth data for specific categories in the training set is incomplete or inaccurately labeled. This affects the model's final generalization ability and recognition performance. Furthermore, the effectiveness of this method is significantly reduced in non-cooperative target scenarios where there is no prior information on pose angles.

[0006] The target recognition method combining deep transfer learning and distribution calibration first pre-trains the network on the source data, and then enhances the classifier training by calibrating the distribution of small sample data and generating new samples, which can improve the recognition rate and solve the overfitting problem. However, this method is not adaptable to extremely small sample scenarios and has poor adaptability to new tasks.

[0007] Therefore, in real-world recognition scenarios, current methods cannot simultaneously overcome the inherent pose sensitivity of HRRP data and the scarcity of training samples, resulting in a sharp decline in recognition performance, or even complete failure. Summary of the Invention

[0008] This invention provides a small-sample radar HRRP target recognition method based on optimal transmission metric, which can solve the above-mentioned technical problems.

[0009] In a first aspect, embodiments of the present invention provide a small-sample radar HRRP target recognition method based on optimal transmission metric, the method comprising: Obtain the HRRP signal of the target to be tested, and obtain the HRPP features of the query set; Based on the optimal transmission scheme, the transmission matrix of the target under test is iteratively solved to obtain the optimal transmission matrix; The transfer matrix is ​​used to characterize the probability that all HRRP features are classified into each category. The HRRP features include the query set HRPP features and the support set HRRP features with known category labels. The category center of each category is updated with the iteration of the transfer matrix, and the category center is the typical feature value of its category. The category of the target to be tested is determined based on the optimal transmission matrix.

[0010] Secondly, embodiments of the present invention provide a small-sample radar HRRP target recognition device based on optimal transmission metric, including a feature extractor, a classification optimization module, and a decision module; The feature extractor is used to obtain the HRRP signal of the target under test and obtain the HRPP features of the query set; The classification optimization module is used to iteratively solve the transmission matrix of the target under test based on the optimal transmission scheme to obtain the optimal transmission matrix; The transfer matrix is ​​used to characterize the probability that all HRRP features are classified into each category. The HRRP features include the query set HRPP features and the support set HRRP features with known category labels. The category center of each category is updated with the iteration of the transfer matrix, and the category center is the typical feature value of its category. The decision module is used to determine the category of the target based on the optimal transmission matrix.

[0011] Thirdly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory is used to store a computer program; the processor can be used to execute a calculator program (instructions) stored in the memory to implement the method of the first aspect described above.

[0012] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed, can implement the method described in the first aspect above.

[0013] It is understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0014] The beneficial effects of this invention compared to existing technologies are as follows: Traditional neural network-based methods treat HRRP features as isolated points in a high-dimensional space, making them insensitive to the overall distribution of features. In contrast, this invention, based on the Wasserstein distance of the optimal transmission scheme, treats the feature set as a probability distribution, transforming the classification decision from traditional nearest-neighbor matching to finding the "minimum cost" distribution alignment. When changes in target pose cause complex movements and deformations of HRRP features in the embedding space, the Wasserstein distance can measure these distribution changes at a lower "cost," thus overcoming the adverse effects of the inherent pose sensitivity of HRRP signals on recognition. Furthermore, the optimal transmission scheme is a non-parametric distance metric that does not require extensive sample training. It only needs to construct the feature space distribution for each category with a very small number of samples beforehand, without needing to learn complex mappings between features like deep networks. Therefore, based on the optimal transmission scheme, accurate target recognition can be achieved even with extremely small sample sizes, such as only 4 or 5 samples per category, improving the adaptability of this invention to new tasks. Finally, by updating the center of each category with the iteration of the transmission matrix, this invention can adaptively adjust the category boundaries, further enhancing its adaptability. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the implementation of a small-sample radar HRRP target recognition method based on optimal transmission metric, provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the implementation of an iterative solution method for the transmission matrix provided by this invention; Figure 3 This is a schematic diagram illustrating an iterative solution of a transmission matrix according to an embodiment of the present invention. Figure 4 A schematic diagram of a small-sample radar HRRP target recognition device based on optimal transmission metric provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a feature extractor provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0016] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0017] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0018] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0019] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0020] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0022] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0023] The method provided in this embodiment of the invention can be applied to electronic devices such as mobile terminals, personal laptops, and supercomputers. This embodiment of the invention does not impose any restrictions on the specific type of electronic device.

[0024] Example 1 Figure 1 The diagram shown illustrates an implementation flowchart of a small-sample radar HRRP target recognition method based on optimal transmission metric, provided by an embodiment of the present invention. As an example and not a limitation, the method may include steps S101-S103, which are described below.

[0025] S101, Obtain the HRRP signal of the target to be tested, and obtain the HRPP features of the query set.

[0026] In one possible implementation, the HRRP signal of the target acquired by the radar can be input into a pre-trained feature extractor to obtain a low-dimensional feature representation (a feature vector) of the target. Then, the feature vector is successively processed by unit vector normalization, zero mean normalization, and QR decomposition, and the decomposed R vector is used as the HRPP feature of the query set.

[0027] Specifically, the query set HRRP features refer to the features obtained based on unlabeled HRRP signal samples to be predicted.

[0028] In one example, unit vector normalization may specifically include: calculating the magnitude of each eigenvector. L2 norm Then normalize it to obtain the normalized feature vector. .

[0029] Normalization can eliminate the inherent amplitude sensitivity of HRRP data, allowing the feature extractor to focus more on the "direction" of the feature rather than its "length", thus enhancing its robustness to changes in factors such as target distance and radar gain.

[0030] In one example, zero-mean means... Subtract the mean of all its dimensions The zero-mean eigenvectors are obtained. .

[0031] Zero-mean processing helps eliminate systematic biases between features, improving the performance and stability of subsequent metric learning.

[0032] In one example, the QR dimensionality reduction decomposition process includes: constructing a matrix from the feature vectors of a batch of HRRP signals. Perform QR decomposition on it, that is ,in It is an orthogonal matrix. It is an upper triangular matrix.

[0033] By preserving the main information of the R matrix, the dimensionality of the feature vector can be effectively reduced without losing too much discriminative information, thereby reducing the complexity of subsequent calculations and minimizing memory usage.

[0034] S102, based on the optimal transmission scheme, the transmission matrix of the target under test is iteratively solved to obtain the optimal transmission matrix.

[0035] For example, the transfer matrix can characterize the probability that all HRPP features are classified into each category.

[0036] Specifically, HRRP features include unlabeled query set HRRP features as well as support set HRRP features with existing category labels.

[0037] For example, the category center of each category is updated with each iteration of the transfer matrix. The category center refers to the typical eigenvalue of that category.

[0038] Specifically, Optimal Transport (OT) theory is a branch of mathematics originating from the classic problem of the 18th-century mathematician Monge. Its core idea is to study how to transform one mass distribution into another target distribution with minimal cost (such as transport distance and energy consumption). In modern mathematics and computer science, OT theory has been extended to measure the distance between two probability distributions. The Wasserstein distance, also known as the bulldozer distance, is an important metric in OT theory. Unlike KL divergence and other metrics that only concern the difference in probability values ​​at corresponding points in the distribution, the Wasserstein distance captures the geometric structure information of the distribution, measuring the "work" required to "transform" one distribution into another. Therefore, it exhibits unique advantages in dealing with complex, non-overlapping distributions.

[0039] S103, Determine the category of the target to be tested based on the optimal transfer matrix.

[0040] Generally, ,in, For the predicted output of the first HRPP features Category tags, Here, it represents the transfer matrix after the last iteration (i.e., the optimal transfer matrix) in the th... OK List the elements.

[0041] Traditional neural network-based methods treat HRRP features as isolated points in a high-dimensional space, making them insensitive to the overall distribution of features. In contrast, this invention, based on the Wasserstein distance of the optimal transmission scheme, treats the feature set as a probability distribution, transforming classification decisions from traditional nearest-neighbor matching to finding the "minimum-cost" distribution alignment. When target pose changes cause complex shifts and deformations in the embedding space of HRRP features, Wasserstein distance can measure these distribution changes at a lower "cost," thus overcoming the adverse effects of the inherent pose sensitivity of HRRP signals on recognition. Furthermore, the optimal transmission scheme is a non-parametric distance metric that requires no large-sample training; it only needs to construct the feature space distribution for each category with a very small number of samples beforehand, unlike deep networks that learn complex mappings between features. Therefore, based on the optimal transmission scheme, accurate target recognition can be achieved even with extremely small sample sizes, such as only 4 or 5 samples per category, improving the adaptability of this invention to new tasks. Finally, by updating the center of each category with the iteration of the transmission matrix, this invention can adaptively adjust the category boundaries, further enhancing its adaptability.

[0042] Example 2 Figure 2The diagram illustrates a flowchart of an iterative solution method for a transfer matrix provided by the present invention. This method is illustrative and not limiting; it can be a possible specific implementation of step S102 described above. The method may include steps S201 to S204, which are described below.

[0043] S201, based on the optimal transmission scheme, determines the transmission matrix for this iteration according to the category center, query set HRPP features, and support set HRRP features after the previous iteration.

[0044] In one example, the unit cost of assigning existing and unlabeled HRPP features to each class can be calculated. Then, with the goal of minimizing the total cost and the allocation entropy, a new loop is started to iteratively solve the transfer matrix after this round of iterations. After finding the minimum value, the new loop ends, and the minimum value obtained in the new loop is taken as the transfer matrix after this round of iterations.

[0045] For example, the transfer matrix after this iteration can satisfy the following formula:

[0046] in, Represents the transmission matrix The Line 1 List the intermediate iteration values ​​of each element in this new loop. Indicates the first HRPP features and The Euclidean distance of the center, i.e. , For the first The category center of each category after the previous iteration. For regularization hyperparameters, For the reason The matrix formed entropy, It is all that satisfy the marginal distribution constraint and The set of joint distribution matrices.

[0047] Optionally, see Figure 3 Before starting target recognition, a small number of HRPP features with labels for each category can be pre-selected to construct a support set. , will the Support sets for each category The average value of the HRPP features of each support set is used as the initial class center for that class of samples, i.e. After obtaining HRPP features of unlabeled samples based on the feature extractor, these features constitute the query set. The transfer matrix after the current iteration is solved iteratively in the new loop based on the Sink-horn algorithm.

[0048] Specifically, the probability that a HRRP feature in the support set is classified into its respective category is 1.

[0049] This invention quantifies the "cost" of assigning any HRRP feature used as a query sample to any category by using the squared Euclidean distance between HRRP features and the center of each category. The lower the cost, the closer the sample is to the center of the category, and the greater the probability that it belongs to that category.

[0050] S202, determine whether the convergence condition is met.

[0051] In one example, if the convergence condition is not met, step S203 can be performed.

[0052] For example, the convergence condition could be that the number of iterations is greater than or equal to the maximum number of iterations, or that the difference in the modulus of the transfer matrix after two adjacent iterations is less than a threshold.

[0053] In another example, if the convergence condition is met, step S204 can be performed.

[0054] S203, update the class center of the previous iteration based on the momentum of the transfer matrix after this iteration, and obtain the class center of the current iteration.

[0055] In one possible implementation, the updated category centers of each category can be determined first based on the transfer matrix after the current iteration, and then the category centers after the previous iteration can be proportionally updated based on the updated category centers of each category to obtain the category centers after the current iteration.

[0056] In one example, the updated category centers can satisfy the following formula:

[0057] in, For the first The updated category center for each category This is the transfer matrix after this iteration. Indicates the use of calculation The function, for The Middle OK List elements, representing the first element. HRPP features (That is, the high-dimensional feature vector extracted from a single unlabeled sample of the target object by the feature extractor and the high-dimensional feature vector extracted from a labeled sample by the feature extractor) are classified into the first category. The probability of each category The total number of categories, The number of HRPP features (i.e., unlabeled samples) in the query set for each category. The support set is a collection of multiple HRPP features that already have category labels. It is a conditional expression representing HRRP features. The corresponding tag is a category. , Belongs to the The total number of labeled sample features for each category.

[0058] In one example, the category centers after this iteration satisfy the following formula:

[0059] in, This serves as the category center after this iteration. For the first The category center of each category after the previous iteration. For the first The updated category center for each category To update the step size.

[0060] Generally, A smaller value can be chosen. When When smaller, category center The updates will be very slow, as they adopt the newly calculated center. At the same time, it retains most of the "inertia" from the previous moment. This smoothing process ensures that the movement trajectory of the category center is more stable.

[0061] After obtaining the category centers after this iteration, we can return to step S201 above and proceed to the next iteration.

[0062] S204 outputs the transmission matrix after this round of iteration as the optimal transmission matrix.

[0063] By updating the category centers in reverse according to the soft labels (i.e., the transfer matrix after this iteration) that contain the distribution structure of the entire query set, the improved optimal transfer method of the present invention (i.e., embodiment 2) can utilize the "bootstrapping" information in the unlabeled data to adaptively adjust the prototypes of each category to a position that better reflects the current distribution of the data to be tested. Example 3 Figure 4 The diagram shown illustrates the structure of a small-sample radar HRRP target recognition device based on optimal transmission metric, provided by an embodiment of the present invention. As an example and not a limitation, the device may include a feature extractor 410, a classification optimization module 420, and a decision module 430.

[0064] For example, feature extractor 410 can be used to acquire the HRRP signal of the target to be tested and obtain the query set HRPP features; classification optimization module 420 can be used to iteratively solve the transfer matrix of the target to be tested based on the optimal transfer scheme to obtain the optimal transfer matrix; wherein, the transfer matrix is ​​used to characterize the probability that all HRRP features are classified into each category, the HRRP features include the query set HRPP features and the support set HRRP features with known category labels, the category center of each category is updated with the iteration of the transfer matrix, and the category center is the typical feature value of its category; decision module 430 can be used to determine the category of the target based on the optimal transfer matrix.

[0065] Traditional neural network-based methods treat HRRP features as isolated points in a high-dimensional space, making them insensitive to the overall distribution of features. In contrast, this invention, based on the Wasserstein distance of the optimal transmission scheme, treats the feature set as a probability distribution, transforming classification decisions from traditional nearest-neighbor matching to finding the "minimum-cost" distribution alignment. When target pose changes cause complex shifts and deformations in the embedding space of HRRP features, Wasserstein distance can measure these distribution changes at a lower "cost," thus overcoming the adverse effects of the inherent pose sensitivity of HRRP signals on recognition. Furthermore, the optimal transmission scheme is a non-parametric distance metric that requires no large-sample training; it only needs to construct the feature space distribution for each category with a very small number of samples beforehand, unlike deep networks that learn complex mappings between features. Therefore, based on the optimal transmission scheme, accurate target recognition can be achieved even with extremely small sample sizes, such as only 4 or 5 samples per category, improving the adaptability of this invention to new tasks. Finally, by updating the center of each category with the iteration of the transmission matrix, this invention can adaptively adjust the category boundaries, further enhancing its adaptability.

[0066] Example 4 Figure 5 The diagram shown illustrates the structure of a feature extractor according to an embodiment of the present invention. By way of example and not limitation, the feature extractor 410 may include three cascaded convolutional blocks and a final fully connected layer.

[0067] In one example, these convolutional blocks can have the same structure, with each block containing a one-dimensional convolutional layer, a ReLU activation function, and a one-dimensional max pooling layer arranged sequentially.

[0068] For example, the convolutional layer has a kernel size of 1×9 and a stride of 1. The max pooling layer has a kernel size of 1×2 and a stride of 2.

[0069] Specifically, the output channels of the three convolutional blocks are different, and can be 32, 64 and 128 respectively.

[0070] In one example, after three convolutional blocks, the feature map is flattened, and its dimensions can then be uniformly mapped to 512 dimensions through a fully connected layer, which is then used as the final feature vector output.

[0071] In one example, to ensure the feature extractor has good generalization ability, it can first be trained on a standard classification task on a large pedestal dataset containing multiple classes. Training uses the cross-entropy loss function and the Adam optimizer. After training, the network parameters are fixed and not updated in subsequent object recognition tasks.

[0072] Example 5 Figure 6 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention. Figure 6 The illustrated electronic device 600 may include: at least one processor 610 ( Figure 6 The diagram shows only one processor, a memory 620, and a computer program 630 stored in the memory 620 and executable on the at least one processor 610, which, when executing the computer program 630, implements the steps in any of the above method embodiments.

[0073] The electronic device 600 may be a robot or other processing device capable of implementing the above methods. This embodiment of the invention does not impose any restrictions on the specific type of electronic device.

[0074] Those skilled in the art will understand that Figure 6 This is merely an example of electronic device 600 and does not constitute a limitation on the electronic device. It may include more or fewer components than shown, or combine certain components, or use different components. For example, electronic device 600 may also include input / output interfaces.

[0075] The processor 610 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASTCs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0076] In some embodiments, the memory 620 may be an internal storage unit, such as a hard disk or RAM. In other embodiments, the memory 620 may be an external storage device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD), or a flash card. Furthermore, the memory 620 may include both internal and external storage units. The memory 620 is used to store the operating system, applications, a boot loader, data, and other programs, such as the program code of the computer program. The memory 620 can also be used to temporarily store data that has been output or will be output.

[0077] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0078] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the functions described above can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0079] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0080] This invention provides a computer program product that, when run on an electronic device, enables the electronic device to perform the steps described in the various method embodiments above.

[0081] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0082] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0083] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A small sample radar HRRP target recognition method based on optimal transmission metric, characterized in that, The method comprises the following steps: acquiring HRRP signals of a target to be detected to obtain query set HRRP features; iteratively solving a transmission matrix of the target to be detected based on an optimal transmission scheme to obtain an optimal transmission matrix; wherein the transmission matrix is used to represent probabilities of all HRRP features being classified into each category, the HRRP features include the query set HRRP features and support set HRRP features with known category labels, and a category center of each category is updated along with the iteration of the transmission matrix, the category center being a typical feature value of the category to which it belongs; determining a category of the target to be detected according to the optimal transmission matrix.

2. The method of claim 1, wherein, Each round of iterative solving of the transmission matrix comprises the following steps: based on the optimal transmission scheme, determining the transmission matrix after this round of iteration according to the category center after the last round of iteration, the query set HRRP features and the support set HRRP features; determining whether a convergence condition is met; if the convergence condition is not met, updating the category center after the last round of iteration according to a momentum of the transmission matrix after this round of iteration to obtain the category center after this round of iteration; if the convergence condition is met, outputting the transmission matrix after this round of iteration as the optimal transmission matrix.

3. The method of claim 2, wherein, The updating of the category center after the last round of iteration according to the momentum of the transmission matrix after this round of iteration to obtain the category center after this round of iteration comprises the following steps: determining updated category centers of all categories according to the transmission matrix after this round of iteration; proportionally updating the category center after the last round of iteration according to the updated category centers to obtain the category center after this round of iteration.

4. The method of claim 3, wherein, The updated category center satisfies the following formula: wherein, is the updated class center of the th class, is the transmission matrix after the current iteration, is a function for calculating , is the th element in the th row and th column, representing the probability that the th HRRP feature is classified into the th class, is the total number of classes, is the number of HRRP features in the query set for each class, is the support set, which is a collection of multiple support set HRRP features with existing class labels, is a conditional expression, representing that the label corresponding to the HRRP feature is the th class, , is the total number of support set HRRP features belonging to the th class.

5. The method of claim 4, wherein, The transmission matrix after this round of iteration satisfies the following formula: wherein, denotes the intermediate iteration value in this round, denotes and the Euclidean distance to the center, is the class center after the last iteration for the th class, is the regularization hyper-parameter, is the entropy of the matrix consisting of is the set of joint distribution matrices satisfying the marginal distribution constraints and is the set of joint distribution matrices satisfying the marginal distribution constraints and 6. The method of claim 3, wherein, The category center after this round of iteration satisfies the following formula: wherein, is the class center after the current iteration, is the class center after the previous iteration for the th class, is the updated class center for the th class, is the update step size.

7. The method of claim 1, wherein, The acquiring of the HRRP signals of the target to be detected to obtain the query set HRRP features comprises the following steps: inputting the HRRP signals of the target to be detected into a feature extractor to obtain a feature vector of the target to be detected; sequentially performing unit vector normalization, zero mean and QR decomposition processing on the feature vector, and taking an R vector after the decomposition as the query set HRRP features.

8. A small sample radar HRRP target recognition device based on optimal transmission metric, characterized in that, The system comprises a feature extractor, a classification optimization module and a decision module. The feature extractor is used to acquire HRRP signals of a target to be detected to obtain query set HRRP features. The classification optimization module is used to iteratively solve a transmission matrix of the target to be detected based on an optimal transmission scheme to obtain an optimal transmission matrix. wherein the transmission matrix is used to represent probabilities of all HRRP features being classified into each category, the HRRP features include the query set HRRP features and support set HRRP features with known category labels, and a category center of each category is updated along with the iteration of the transmission matrix, the category center being a typical feature value of the category to which it belongs. The decision module is used to determine a category of the target according to the optimal transmission matrix.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program, which is executed by the electronic device, implements the method according to any one of claims 1 to 7.

Images