Robust Identification Method for Radar Target Multi-Mode Data under Resource Constrained Conditions

Abstract

The present invention discloses a robust recognition method for radar target multi-modal data under resource-constrained conditions, including: inputting the radar multi-modal data to be recognized into a trained radar multi-modal data recognition model; wherein, the modalities of the data include: high-resolution one-dimensional range image modality, narrowband modulation spectrum modality, and status information sequence modality; the model includes a single-modal mixture of experts learning module and a multi-modal feature confidence fusion module; obtaining the evidence, uncertainty, and class probability corresponding to each modality data of the radar according to the trained multi-modal feature confidence fusion module and the features corresponding to each modality data of the radar obtained according to the trained single-modal mixture of experts learning module, and performing confidence fusion, and obtaining the recognition result according to the obtained target confidence fusion result. The present invention fully excavates the features of each modality of radar data, improves the accuracy of the description of radar multi-modal data, and enhances the reliability and accuracy of target recognition through confidence fusion.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar, and particularly relates to a robust recognition method for radar target multi-modal data under resource-constrained conditions. Background Art

[0002] The high-resolution range profile (HRRP) of a radar reflects the distribution of the main scattering centers of a target along the radar line of sight at a certain radar viewing angle, and contains the geometric characteristics and electromagnetic scattering characteristics of the main structure of the target. It is widely used in the field of radar target recognition. However, limited by system resources, there is a problem of reduced information due to insufficient bandwidth in HRRP. In addition, due to the complexity of the actual application environment, HRRP may also face problems affected by noise and interference, making it difficult to robustly identify targets. Some existing radar target recognition technologies fuse HRRP and synthetic aperture radar (SAR) data, but do not consider the description of the physical characteristics of targets by different modalities of radar data, resulting in poor radar target recognition effects. Summary of the Invention

[0003] To solve the above problems existing in the prior art, the present invention provides a robust recognition method for radar target multi-modal data under resource-constrained conditions.

[0004] The technical problems to be solved by the present invention are realized through the following technical solutions:

[0005] In a first aspect, the present invention provides a robust recognition method for radar target multi-modal data under resource-constrained conditions, and the method includes:

[0006] Inputting the radar multi-modal data to be recognized into a trained radar multi-modal data recognition model; wherein, the modalities of the radar multi-modal data to be recognized include: high-resolution range profile modality, narrowband modulation spectrum modality, and status information sequence modality; the radar multi-modal data recognition model includes a single-modal mixture of experts learning module and a multi-modal feature confidence fusion module;

[0007] Obtaining the features corresponding to each modality data of the radar in the radar multi-modal data to be recognized according to the trained single-modal mixture of experts learning module;

[0008] Obtaining the evidence, uncertainty, and class probability corresponding to each modality data of the radar in the radar multi-modal data to be recognized according to the trained multi-modal feature confidence fusion module and the features corresponding to each modality data of the radar in the radar multi-modal data to be recognized; wherein, the evidence corresponding to each modality data of the radar represents the degree of support for each category to which the radar multi-modal data to be recognized belongs;

[0009] Perform confidence fusion based on the evidence, uncertainty, and class probabilities corresponding to each modality data of the radar in the radar multi-modal data to be recognized, and obtain the target confidence fusion result;

[0010] Obtain the recognition result of the radar multi-modal data to be recognized according to the target confidence fusion result.

[0011] Optionally, the training process of the radar multi-modal data recognition model includes:

[0012] The training process of the radar multi-modal data recognition model includes:

[0013] Construct a training data set; wherein, each sample in the training data set is labeled with a corresponding true class, and the modalities of each sample include: high-resolution one-dimensional range image modality, narrowband modulation spectrum modality, and status information sequence modality;

[0014] Train the radar multi-modal data recognition model according to the training data set and a preset loss function until a preset condition is reached, and obtain the trained radar multi-modal data recognition model.

[0015] Optionally, the preset loss function includes: load balancing loss; the load balancing loss is expressed as follows:

[0016] ;

[0017] Wherein, represents the load balancing loss corresponding to modality , represents the number of experts corresponding to the modality , represents the token ratio assigned to the expert in the single-modal mixture of experts learning module in the modality , represents the average probability score assigned to the expert in the modality .

[0018] Optionally, the preset loss function further includes: confidence fusion loss; the confidence fusion loss is expressed as follows:

[0019] ;

[0020] Wherein, represents the confidence fusion loss, represents the cross-entropy loss, represents the regularization loss, represents the sample The Dirichlet distribution parameter vector after confidence fusion.

[0021] Optionally, the cross-entropy loss is expressed as follows:

[0022] ;

[0023] where is the true label of the sample in the th class in one-hot format, represents the probability that the sample is predicted to be in the th class, is a D-dimensional polynomial function, represents the number of classes, represents the value of the sample in the th class, represents the class probability vector obtained with as the Dirichlet parameter, represents the digamma function, represents the Dirichlet strength of the sample .

[0024] Optionally, the regularization loss is expressed as follows:

[0025] ;

[0026] where represents the all-ones vector, represents the Kullback-Leibler divergence, is the Dirichlet parameter vector after removing non-misleading evidence from , is the true label of the sample in one-hot format, is the probability density function of the Dirichlet distribution, is the gamma function, represents the Dirichlet parameter after removing non-misleading evidence for the th class in

[0027] Optionally, the target confidence fusion result is expressed as follows:

[0028] ;

[0029] where represents the target confidence fusion result, Represents the uncertainty of the high-resolution one-dimensional range image modality, Represents the uncertainty of the narrowband modulation spectrum modality, Represents the uncertainty of the state information sequence modality, Represents the evidence of the high-resolution one-dimensional range image modality, Represents the evidence of the narrowband modulation spectrum modality, Represents the evidence of the state information sequence modality.

[0030] Optionally, obtaining the evidence, uncertainty, and class probability corresponding to each radar modality data in the to-be-identified radar multi-modal data according to the trained multi-modal feature confidence fusion module and the features corresponding to each radar modality data in the to-be-identified radar multi-modal data includes:

[0031] After passing the features corresponding to each radar modality data in the to-be-identified radar multi-modal data through the linear layer and Softplus in the multi-modal feature confidence fusion module, obtaining the evidence corresponding to each radar modality data;

[0032] Calculating the uncertainty and class probability corresponding to each radar modality data in the to-be-identified radar multi-modal data according to the evidence corresponding to each radar modality data.

[0033] The technical solution provided by the embodiments of the present invention may include the following beneficial effects:

[0034] The present invention comprehensively utilizes the high-resolution one-dimensional range image modality, narrowband Doppler modulation spectrum modality, and state information sequence modality of the radar, fully excavates the features of radar data from multiple modalities and multiple dimensions, and improves the accuracy of the attribute description of the to-be-identified radar multi-modal data; the present invention also considers the possible modal quality uncertainty in the radar multi-modal data, enhances the processing ability of the model by using a single-modal mixture of experts learning module in the model structure design to select different expert networks to process data of different modalities, and designs a multi-modal feature confidence fusion module based on modal uncertainty in the fusion strategy to perform confidence fusion on each radar modality data, effectively improving the reliability and accuracy of radar multi-modal data fusion target recognition.

[0035] The following will further describe the present invention in detail with reference to the accompanying drawings and embodiments. Description of the Drawings

[0036] Figure 1 Is a flowchart of a method for robust recognition of radar target multi-modal data under resource-constrained conditions provided by an embodiment of the present invention;

[0037] Figure 2 Is a schematic diagram of a single-modal mixture of experts learning module provided by an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of a multi-modal feature confidence fusion module provided by an embodiment of the present invention. Specific implementation manners

[0039] The present invention will be further described in detail below with reference to specific embodiments, but the implementation manners of the present invention are not limited thereto.

[0040] Figure 1 It is a flowchart of a method for robust recognition of radar target multi-modal data under resource-constrained conditions provided by an embodiment of the present invention. As Figure 1 shown, the method may include the following steps:

[0041] S101. Input the radar multi-modal data to be recognized into the trained radar multi-modal data recognition model; wherein, the modalities of the radar multi-modal data to be recognized include: high-resolution one-dimensional range image modality, narrowband modulation spectrum modality, and state information sequence modality; the radar multi-modal data recognition model includes a single-modal mixture of experts learning module and a multi-modal feature confidence fusion module.

[0042] Exemplarily, the radar multi-modal data to be recognized may be referred to as follows: the radar multi-modal data to be recognized may be three types of aircraft target data, composed of the data of aircraft A, aircraft B, and aircraft C. Each data includes a radar one-dimensional high-resolution range image with a dimension of 1×256; a narrowband modulation spectrum with a dimension of 1×1024; a state information sequence, which contains 10 moments, and each moment contains target distance, azimuth, pitch, attitude, height, speed, and radar cross section, with a dimension of 10×7.

[0043] S102. Obtain the features corresponding to each radar modality data in the radar multi-modal data to be recognized according to the trained single-modal mixture of experts learning module.

[0044] It can be understood that each modality has a corresponding single-modal mixture-of-experts learning module. Therefore, the single-modal mixture-of-experts learning module can include: a range profile single-modal mixture-of-experts learning module, a modulation spectrum single-modal mixture-of-experts learning module, and a state information sequence single-modal mixture-of-experts learning module. The range profile single-modal mixture-of-experts learning module includes: a range profile preprocessing module, a range profile feature extraction module, and a range profile feature mixture-of-experts learning module. The input of the range profile single-modal mixture-of-experts learning module is a high-resolution one-dimensional range profile, and the output is the extracted range profile features and the range profile mixture-of-experts load balancing loss; the modulation spectrum single-modal mixture-of-experts learning module includes: a modulation spectrum preprocessing module, a modulation spectrum feature extraction module, and a modulation spectrum feature mixture-of-experts learning module. The input of the modulation spectrum single-modal mixture-of-experts learning module is a narrowband modulation spectrum, and the output is the extracted modulation spectrum features and the modulation spectrum mixture-of-experts load balancing loss; the state information sequence single-modal mixture-of-experts learning module includes: a state information sequence preprocessing module, a state information sequence feature extraction module, and a state information sequence feature mixture-of-experts learning module. The input of the state information sequence single-modal mixture-of-experts learning module is a state information sequence, and the output is the extracted state information sequence features and the state information sequence mixture-of-experts load balancing loss.

[0045] Specifically, Figure 2 is a schematic diagram of a single-modal mixture-of-experts learning module provided by an embodiment of the present invention. As Figure 2 shown, taking the radar multi-modal data to be recognized as an example, the high-resolution one-dimensional range profile in the radar multi-modal data to be recognized is input into the range profile preprocessing module. The preprocessing can be normalized and centroid-aligned using the L2 norm. Then it is input into the range profile feature extraction module. The feature extraction can use a 3-layer one-dimensional convolutional network, with each layer containing 1 convolutional layer, 1 batch normalization layer, and 1 ReLU activation layer. The number of convolutional channels is (8, 16, 32) respectively, the convolutional kernel size is 3, the stride is 2, and the padding is 1. Then the features extracted by the range profile feature extraction module are input into the range profile feature mixture-of-experts learning module. The range profile feature mixture-of-experts learning module first adds learnable parameters with a dimension of 128 and the same length as the range profile feature token sequence as position encoding. After adding the range profile feature token sequence and the position encoding, it passes through 1 self-attention layer (including 3 linear layers, with the number of nodes being (256, 128)), 1 layer normalization layer, 1 expert probability mapping layer (including 1 linear layer, with the number of nodes being (128, 4)), 1 Softmax layer, 1 mixture-of-experts layer (including 4 feed-forward experts, each expert containing 1 linear layer, with the number of nodes being (128, 512)), 1 GeLU activation layer, 1 linear layer (with the number of nodes being (512, 128)), 1 layer normalization layer, and an average layer to obtain the range profile feature output, with the number of nodes being (1, 128). In this embodiment, the number of experts selected is 1.

[0046] Input the narrowband modulation spectrum in the radar multi-modal data to be recognized into the modulation spectrum preprocessing module. The preprocessing adopts clutter removal, translation of the main component, and normalization, and then inputs it into the modulation spectrum feature extraction module. The feature extraction adopts a 3-layer one-dimensional convolutional network. Each layer contains 1 convolutional layer, 1 batch normalization layer, 1 ReLU activation layer, and 1 max pooling layer. The number of convolutional channels is (8, 16, 32) respectively, the kernel size of the convolution is 3, the stride is 2, the padding is 1, and the pooling stride is 2. Then, input the features extracted by the modulation spectrum feature extraction module into the modulation spectrum feature mixture of experts learning module. The modulation spectrum feature mixture of experts learning module first adds learnable parameters with a dimension of 128 and the same length as the modulation spectrum feature token sequence as the position encoding. After adding the modulation spectrum feature token sequence and the position encoding, it passes through 1 self-attention layer (including 3 linear layers, the number of nodes is (128, 128)), passes through 1 layer normalization layer, passes through 1 expert probability mapping layer (including 1 linear layer, the number of nodes is (128, 4)), 1 Softmax layer, 1 mixture of experts layer (including 4 feed-forward experts, each expert includes 1 linear layer, the number of nodes is (128, 512)), 1 GeLU activation layer, 1 linear layer (the number of nodes is (512, 128)), passes through 1 layer normalization layer, and passes through an average layer to obtain the modulation spectrum feature output (the number of nodes is (1, 128)). The number of experts selected in this embodiment is 1.

[0047] Input the state information sequence in the radar multi-modal data to be recognized into the state information preprocessing module. Rearrange the data format and normalize the state information sequence, and then input it into the state information feature extraction module. The feature extraction adopts a 3-layer bidirectional long short-term memory network. Each layer contains 1 forward long short-term memory layer and 1 backward long short-term memory layer, and the hidden layer dimension is 64. Then input it into the state information feature mixture of experts learning module. The state information feature mixture of experts learning module first adds learnable parameters with a dimension of 128 and the same length as the token of the state information feature sequence as the position encoding. After adding the token sequence of the state information feature sequence and the position encoding, it passes through 1 self-attention layer (including 3 linear layers, the number of nodes is (128, 128)), passes through 1 layer normalization layer, passes through 1 expert probability mapping layer (including 1 linear layer, the number of nodes is (128, 4)), 1 Softmax layer, 1 mixture of experts layer (including 4 feed-forward experts, each expert includes 1 linear layer, the number of nodes is (128, 512)), 1 GeLU activation layer, 1 linear layer (the number of nodes is (512, 128)), passes through 1 layer normalization layer, and passes through an average layer to obtain the state information feature output (the number of nodes is (1, 128)). The number of experts selected in this embodiment is 1.

[0048] S103. Obtain the evidence, uncertainty, and class probability corresponding to each radar modality data in the radar multi-modal data to be recognized based on the trained multi-modal feature confidence fusion module and the features corresponding to each radar modality data in the radar multi-modal data to be recognized; wherein, the evidence corresponding to each radar modality data represents the degree of support for each category to which the radar multi-modal data to be recognized belongs.

[0049] It can be understood that the uncertainty corresponding to each radar modality data in the radar multi-modal data to be recognized represents the reliability of the result; the class probability corresponding to each radar modality data in the radar multi-modal data to be recognized represents the confidence in the recognition result of the category to which each radar modality data in the radar multi-modal data to be recognized belongs.

[0050] Optionally, Figure 3 is a schematic diagram of a multi-modal feature confidence fusion module provided by an embodiment of the present invention. S103 may include:

[0051] After passing the features corresponding to each radar modality data in the radar multi-modal data to be recognized through the linear layer and Softplus in the multi-modal feature confidence fusion module, obtain the evidence corresponding to each radar modality data;

[0052] Calculate the uncertainty and class probability corresponding to each radar modality data in the radar multi-modal data to be recognized according to the evidence corresponding to each radar modality data.

[0053] It can be understood that after passing the features corresponding to each radar modality data in the radar multi-modal data to be recognized through the linear layer and Softplus respectively, the evidence corresponding to each radar modality data is obtained. The number of nodes in the three linear layers is (128, 3). Calculate the Dirichlet distribution parameters, class beliefs, uncertainty, and class probability. In this embodiment, the number of classes is , for example can be 3, and the class beliefs , uncertainty and class probability corresponding to each modality are calculated as follows:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] wherein, modality , Represents the high-resolution one-dimensional range image mode, Represents the narrowband modulation spectrum mode, Represents the state information sequence mode, is the mode of the Dirichlet distribution parameter, is the mode of the Dirichlet intensity, is the mode of the evidence.

[0060] S104. Perform confidence fusion based on the evidence, uncertainty, and class probability corresponding to each radar mode data in the radar multi-mode data to be recognized, and obtain the target confidence fusion result.

[0061] Optionally, the target confidence fusion result is expressed as follows:

[0062] ;

[0063] wherein, represents the target confidence fusion result, represents the uncertainty of the high-resolution one-dimensional range image mode, represents the uncertainty of the narrowband modulation spectrum mode, represents the uncertainty of the state information sequence mode, represents the evidence of the high-resolution one-dimensional range image mode, represents the evidence of the narrowband modulation spectrum mode, represents the evidence of the state information sequence mode.

[0064] S105. Obtain the recognition result of the radar multi-mode data to be recognized according to the target confidence fusion result.

[0065] It can be understood that, according to the target confidence fusion result, the confidence of the radar multi-mode data to be recognized belonging to each category is obtained, and the category with the highest confidence is used as the category of the radar multi-mode data to be recognized, that is, the recognition result of the radar multi-mode data to be recognized.

[0066] Optionally, the training process of the radar multi-mode data recognition model includes:

[0067] Construct a training data set; wherein, each sample in the training data set is labeled with the corresponding true category, and the modes of each sample include: high-resolution one-dimensional range image mode, narrowband modulation spectrum mode, and state information sequence mode;

[0068] Train the radar multi-mode data recognition model according to the training data set and the preset loss function until the preset conditions are met, and obtain the trained radar multi-mode data recognition model.

[0069] Exemplarily, the total number of samples in the training dataset can be 3000. Each sample contains 1 radar high-resolution one-dimensional range image, 1 narrowband modulation spectrum, and 1 target state information sequence. Each target state information sequence contains data for 10 moments, and each moment contains the target distance, azimuth, pitch, attitude, altitude, speed, and radar cross-sectional area. The training dataset contains 3 types of targets, among which there are 1000 samples of aircraft A, 1000 samples of aircraft B, and 1000 samples of aircraft C. The total number of samples in the test dataset is 1500, and the composition of each sample is the same as that of the training dataset, among which there are 500 samples of aircraft A, 500 samples of aircraft B, and 500 samples of aircraft C. The test dataset is used to test the performance of the radar multi-modal data recognition model. The preset condition is that the sum of each loss reaches the minimum during the limited number of iterative training processes. Exemplarily, when training the radar multi-modal data recognition model, the batch size within each round can be 64, the initial learning rate is 1e-4, the optimizer is AdamW, and the total number of iterative rounds is set to 100.

[0070] Optionally, the preset loss function includes: load balancing loss; the load balancing loss is expressed as follows:

[0071] ;

[0072] Among them, represents the load balancing loss corresponding to the modality , represents the number of experts corresponding to the modality , represents the token ratio assigned to the expert in the single-modal mixture of experts learning module in the modality , represents the average probability score assigned to the expert in the modality .

[0073] ;

[0074] Among them, represents the number of feature tokens corresponding to the modality , is an indicator function, represents any batch of training data in the training dataset, represents the training data of the modality , represents the probability of selecting each expert obtained according to ;

[0075] ;

[0076] Among them, represents according to The obtained selected expert probability

[0077] Optionally, the preset loss function further includes: a confidence fusion loss; the confidence fusion loss is expressed as follows:

[0078] ;

[0079] wherein represents the confidence fusion loss, represents the cross-entropy loss, represents the regularization loss, represents the sample Dirichlet distribution parameter vector after confidence fusion

[0080] Optionally, the cross-entropy loss is expressed as follows:

[0081] ;

[0082] wherein is the true label of the sample in the th class in one-hot format, represents the sample being predicted as the th class probability, is dimensional polynomial function, represents the number of classes, represents the value of the sample in the th class, represents the class probability vector obtained with as the Dirichlet parameter, represents the digamma function, represents the sample Dirichlet intensity

[0083] The regularization loss is expressed as follows:

[0084] ;

[0085] wherein represents the all-ones vector, represents the Kullback-Leibler divergence, is the Dirichlet parameter vector after removing non-misleading evidence from , is the true label of the sample in one-hot format, is the probability density function of the Dirichlet distribution is the gamma function, denotes the Dirichlet parameter after removing non-misleading evidence in the th class.

[0086] The effectiveness of the present invention is illustrated by the following simulation experiments.

[0087] Specifically, the hardware platform for the simulation experiment is as follows: the processor is an Intel(R) i7 CPU with a main frequency of 3.20 GHz, the memory capacity is 64 GB, the graphics card is an Nvidia RTX 3090, and the video memory capacity is 24 GB. The software platform for the simulation experiment is: Python 3.8, Pytorch 2.0.

[0088] In the simulation experiment of the present invention, the present invention and a prior art are used to conduct target recognition experiments on three types of aircraft, and the recognition rates are compared. To simulate the disturbance situation in the actual scenario, Gaussian noise with a standard deviation of 1 is added to any modality of 50% of the test data set in the experiment. The prior art refers to the paper "Deep Residual Learning for Image Recognition", which applies the ResNet network to the recognition model. The original network is designed for two-dimensional image data. In this embodiment, its main structure is retained and the two-dimensional operations are modified to one-dimensional operations to adapt to HRRP data. The recognition rate is used to evaluate the performance of the recognition model in the simulation experiment, which is defined as the ratio of the number of correctly recognized samples to the total number of samples in the test data set. The higher the recognition rate, the more correctly recognized samples, and the better the recognition performance of the model.

[0089] Table 1

[0090]

[0091] By comparing the recognition rates of the present invention and the prior art, it can be seen that the present invention effectively improves the radar target recognition performance in the disturbance scenario.

[0092] The present invention comprehensively utilizes the radar high-resolution one-dimensional range image modality, narrowband Doppler modulation spectrum modality, and state information sequence modality, fully excavates the features of radar data from multiple modalities and dimensions, and improves the accuracy of the attribute description of the radar multi-modal data to be recognized; the present invention also considers the possible modality quality uncertainty in the radar multi-modal data, enhances the processing ability of the model by using a single-modal mixture of experts learning module in the model structure design, so as to select different expert networks to process data of different modalities, and designs a multi-modal feature confidence fusion module based on modality uncertainty in the fusion strategy to perform confidence fusion on radar multi-modal data, effectively improving the reliability and accuracy of radar multi-modal data fusion target recognition.

[0093] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc., mean that the specific features or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification.

[0094] Although the present invention has been described herein in connection with various embodiments, however, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the accompanying drawings and the disclosure. In the description of the present invention, the term "comprising" does not exclude other components or steps, the word "a" or "an" does not exclude a plurality of cases, and the meaning of "a plurality" is two or more, unless otherwise specifically defined. In addition, certain measures are described in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0095] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A robust identification method for radar target multi-mode data under resource-constrained conditions, characterized in that: The method comprises: Inputting the radar multimodal data to be identified into the trained radar multimodal data recognition model; wherein the modes of the radar multimodal data to be identified include: high-resolution one-dimensional range image mode, narrow-band modulation spectrum mode and state information sequence mode; the radar multimodal data recognition model includes a single-modal hybrid expert learning module and a multimodal feature confidence fusion module; the trained radar multimodal data recognition model is obtained by training according to a training data set and a preset loss function; Obtaining features corresponding to each modal data of the radar in the multimodal data of the radar to be identified according to the trained single-modal hybrid expert learning module; Obtaining evidence, uncertainty and category probability corresponding to each modal data of radar in the multimodal data of radar to be identified according to the trained multimodal feature confidence fusion module and the features corresponding to each modal data of radar in the multimodal data of radar to be identified; wherein the evidence corresponding to each modal data of radar represents the support degree for each category to which the multimodal data of radar to be identified belongs; According to the evidence, uncertainty and category probability corresponding to each modal data of the radar in the multimodal data of the radar to be identified, confidence fusion is performed to obtain a target confidence fusion result; Obtaining an identification result of the radar multimodal data to be identified according to the target confidence fusion result; The preset loss function includes: load balancing loss; the load balancing loss is expressed as follows: ； in, Representing modality The corresponding load balancing loss is Indicates the mode The corresponding number of experts, Indicates that in the modal Assigned to the experts in the unimodal hybrid expert learning module The token ratio, Indicates that in the modal Assigned to the expert The average probability score of The preset loss function also includes: confidence fusion loss; the confidence fusion loss is expressed as follows: ； in, represents the confidence fusion loss, represents the cross entropy loss, represents the regularization loss, Representation sample Parameter vector of the Dirichlet distribution after confidence fusion.

2. The method for robust identification of radar target multi-mode data under resource-constrained conditions according to claim 1 is characterized in that: The training process of the radar multimodal data recognition model includes: Constructing a training data set; wherein each sample in the training data set is labeled with a corresponding real category, and the modality of each sample includes: a high-resolution one-dimensional range image modality, a narrow-band modulation spectrum modality, and a state information sequence modality; The radar multimodal data recognition model is trained according to the training data set and a preset loss function until a preset condition is met, thereby obtaining the trained radar multimodal data recognition model.

3. The method for robust identification of radar target multi-mode data under resource-constrained conditions according to claim 1, characterized in that: The cross entropy loss is expressed as follows: ； in, For the sample No. True labels in one-hot format, Representation sample Predicted to be The probability of the class, for dimensional polynomial function, represents the number of categories, express The sample described in No. The value of the class, Indicates is the class probability vector obtained by Dirichlet parameters, represents the digamma function, Indicates the sample The Dirichlet intensity of .

4. The method for robust identification of radar target multi-mode data under resource-constrained conditions according to claim 3 is characterized in that: The regularization loss is expressed as follows: ； in, represents a vector of all 1s, represents the Kullback-Leibler divergence, For The Dirichlet parameter vector after removing non-misleading evidence, For sample The true label in one-hot format, is the probability density function of the Dirichlet distribution, is the gamma function, express Middle Dirichlet parameter after removing non-misleading evidence.

5. The method for robust identification of radar target multi-mode data under resource-constrained conditions according to claim 1, characterized in that: The target confidence fusion result is expressed as follows: ； in, represents the target confidence fusion result, represents the uncertainty of the high-resolution one-dimensional range image mode, represents the uncertainty of the narrowband modulation spectrum mode, represents the uncertainty of the state information sequence mode, represents the evidence of the high-resolution one-dimensional range image modality, represents evidence of the narrowband modulation spectrum modality, Evidence representing the modality of the state information sequence.

6. The method for robust identification of radar target multi-mode data under resource-constrained conditions according to claim 1, characterized in that: The obtaining of evidence, uncertainty and category probability corresponding to each modal data of the radar in the multimodal data of the radar to be identified based on the trained multimodal feature confidence fusion module and the features corresponding to each modal data of the radar in the multimodal data of the radar to be identified includes: The features corresponding to each modal data of the radar in the radar multimodal data to be identified are subjected to the linear layer and Softplus in the multimodal feature confidence fusion module to obtain the evidence corresponding to each modal data of the radar; The uncertainty and category probability corresponding to each radar modal data in the radar multimodal data to be identified are calculated according to the evidence corresponding to each radar modal data.

Images