Radar HRRP Target Recognition Method, Device and Electronic Equipment Based on Attitude Angle Auxiliary Information

Abstract

The present invention provides a radar HRRP target recognition method, device and electronic device based on attitude angle auxiliary information, relating to the field of radar technology, aiming to solve the problem that the attitude angle labels of sample targets in the training set are available in practical applications, but the attitude angle labels of the test set are missing. The method includes: inputting radar HRRP data into the parameter sharing network of the target recognition model to obtain feature information; inputting the feature information into the first specific task network to obtain the recognition result of the target category; the target recognition model is trained based on sample radar HRRP data, category labels and attitude angle labels of sample targets. In the training of the present invention, the output information of the parameter sharing network is jointly adopted by the two specific task networks, and this output information is obtained based on the input sample radar HRRP data, without the need to input the attitude angle of the target to be recognized, so there is no need to input the attitude angle in the test stage either, improving the performance of the target recognition model.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to a radar HRRP target recognition method, device, and electronic device based on attitude angle auxiliary information. Background Art

[0002] Radar High Resolution Range Profile (HRRP) has the characteristics of low dimension, easy acquisition, and easy processing, and is a commonly used tool for radar target recognition. HRRP reflects the position distribution of the target's Radar Cross Section (RCS) in the radar line-of-sight direction, and contains the structural and position information between each scatterer. Using the HRRP data of the target, an end-to-end deep network can be trained to achieve feature extraction and class prediction of the target to be recognized, and then a response plan can be executed based on the class prediction.

[0003] In the related art, usually the attitude angle corresponding to the sample target is used as auxiliary information, and both the sample HRRP data and the attitude angle corresponding to the sample target are used as input data of the target recognition model for training.

[0004] However, in the above related art, when training the target recognition model, since both inputs need to be used as input data, the training set and the test set are required to have attitude angles. However, in the real scenario, the echo received by the radar does not contain the corresponding attitude angle. Therefore, in the test stage, the input corresponding to the attitude angle will be set to null, which will lead to the lack of attitude angle information in the test stage, and thus reduce the performance of the finally trained target recognition model. Summary of the Invention

[0005] The present invention provides a radar HRRP target recognition method, device, and electronic device based on attitude angle auxiliary information to solve the defect in the prior art that the performance of the finally trained target recognition model is reduced.

[0006] The present invention provides a radar HRRP target recognition method based on attitude angle auxiliary information, including the following steps.

[0007] Obtain the radar HRRP data of the target to be recognized;

[0008] Input the radar HRRP data into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network;

[0009] Input the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network;

[0010] Among them, the target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the class label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. Both the first specific task network and the second specific task network are connected to the parameter sharing network. The first specific task network is used to output a predicted class, and the second specific task network is used to output a predicted attitude angle.

[0011] According to a radar HRRP target recognition method based on attitude angle auxiliary information provided by the present invention, the target recognition model is trained based on the following method:

[0012] Obtain the sample radar HRRP data of the sample target;

[0013] Input the sample radar HRRP data into the parameter sharing network of the initial target recognition model to obtain the sample feature information output by the parameter sharing network;

[0014] Input the sample feature information into the first specific task network and the second specific task network of the initial target recognition model respectively to obtain the predicted class output by the first specific task network and the predicted attitude angle output by the second specific task network;

[0015] Based on the predicted class, the class label of the sample target, the predicted attitude angle, and the attitude angle label corresponding to the sample target, determine the total loss function;

[0016] Based on the total loss function, iteratively adjust the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network to obtain the target recognition model.

[0017] According to a radar HRRP target recognition method based on attitude angle auxiliary information provided by the present invention, determining the total loss function based on the predicted class, the class label of the sample target, the predicted attitude angle, and the attitude angle label corresponding to the sample target includes:

[0018] Based on the predicted class and the class label of the sample target, determine the recognition task loss function;

[0019] Based on the predicted attitude angle and the attitude angle label corresponding to the sample target, determine the angle estimation task loss function;

[0020] Based on the following formula (1), determine the total loss function:

[0021] (1)

[0022] in, represents the total loss function, represents the network parameters of the target recognition model, and represents the trainable parameters, represents the angle estimation task loss function, Represents the recognition task loss function.

[0023] According to a radar HRRP target recognition method based on attitude angle auxiliary information provided by the present invention, obtaining sample radar HRRP data of the sample target includes:

[0024] The optimization problem is constructed based on the following formula (2):

[0025] (2)

[0026] Among them, the constraints are , Indicates the total number of sample targets, represents the network parameters of the target recognition model, and represents the trainable parameters, represents the sample selection threshold, Indicates the The control parameters of the sample target, It is expressed by the following formula (3):

[0027] (3)

[0028] represents the regularization term, , Indicates the Multi-task joint loss of sample targets,

[0029] , represents the weight of the angle estimation task, Indicates the The angle estimation task loss function corresponding to the sample target is: Indicates the The pose angle labels corresponding to the sample targets, Indicates the The predicted pose angle corresponding to the sample target, represents the network parameters of the second task-specific network, Indicates the The recognition task loss function corresponding to the sample target is: Indicates the A sample target corresponds to a classification label, indicating the predicted category corresponding to the th sample target;

[0030] Solve the optimization problem to obtain the network parameters of the target recognition model .

[0031] According to a radar HRRP target recognition method based on attitude angle auxiliary information provided by the present invention, determining an angle estimation task loss function based on the predicted attitude angle and the attitude angle label corresponding to the sample target includes:

[0032] Determine the interval width based on the following formula (4):

[0033] (4)

[0034] where represents the interval width, represents the maximum angle within the angle space range, represents the minimum angle within the angle space range, represents the total number of preset intervals, represents the mapping function;

[0035] Determine the discrete angle corresponding to the predicted attitude angle of the sample target based on the following formula (5):

[0036] (5)

[0037] where represents the predicted attitude angle

[0038] corresponding discrete angle;

[0039] (6)

[0040] where represents the metric after distance penalty, represents the serial number of the interval where the predicted attitude angle is located, represents the weight;

[0041] Determine the distance penalty error of the discrete angle relative to different intervals based on the following formula (7):

[0042] (7)

[0043] Among them, represents the discrete angle with respect to the distance penalty error of different intervals, represents the serial number of the k-th interval;

[0044] Based on the distance penalty error of the discrete angle, the predicted pose angle, and the pose angle label corresponding to the sample target, determine the angle estimation task loss function.

[0045] According to a radar HRRP target recognition method based on pose angle auxiliary information provided by the present invention, determining the angle estimation task loss function based on the distance penalty error of the discrete angle, the predicted pose angle, and the pose angle label corresponding to the sample target includes:

[0046] Perform mapping processing on the distance penalty error of the discrete angle to obtain the probability density of the discrete angle in different intervals;

[0047] Determine the angle estimation task loss function based on the following formula (8):

[0048] (8)

[0049] Among them, represents the angle estimation task loss function, represents the total number of sample targets, represents the probability density of the discrete angle of the -th sample target in different intervals, represents the vector of the pose angle label of the -th sample target, represents the cosine weight of the , represents the pose angle label of the

[0050] According to a radar HRRP target recognition method based on pose angle auxiliary information provided by the present invention, the parameter sharing network includes a first convolution module, a second convolution module, a third convolution module, and a first fully connected layer connected in sequence. The first convolution module, the second convolution module, and the third convolution module each include a convolution layer, a batch normalization layer, a first activation function layer, and a max pooling layer connected in sequence. The first specific task network and the second specific task network each include a second fully connected layer, a second activation function layer, and a third fully connected layer connected in sequence.

[0051] The present invention also provides a radar HRRP target recognition device based on pose angle auxiliary information, including:

[0052] An acquisition unit, configured to acquire radar HRRP data of a target to be recognized;

[0053] An extraction unit, configured to input the radar HRRP data into a parameter sharing network of a target recognition model to obtain feature information output by the parameter sharing network;

[0054] An identification unit, configured to input the feature information into a first specific task network of the target recognition model to obtain an identification result of the target category output by the first specific task network;

[0055] Wherein, the target recognition model is obtained by training an initial target recognition model based on sample radar HRRP data of a sample target, a category label of the sample target, and a pose angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is configured to output a predicted category, and the second specific task network is configured to output a predicted pose angle.

[0056] 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 radar HRRP target recognition based on pose angle auxiliary information as described in any one of the above is implemented.

[0057] 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 radar HRRP target recognition based on pose angle auxiliary information as described in any one of the above is implemented.

[0058] 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 radar HRRP target recognition based on pose angle auxiliary information as described in any one of the above is implemented.

[0059] The radar HRRP target recognition method, device, and electronic device based on attitude angle auxiliary information provided by the present invention input the obtained radar HRRP data of the target to be recognized into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network, and then input the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network. Among them, the target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted attitude angle. It can be seen that when training the target recognition model in the present invention, the first specific task network and the second specific task network jointly adopt the output information of the parameter sharing network, and this output information is obtained based on the input sample radar HRRP data, and it is not necessary to input the attitude angle of the target to be recognized, but only use the attitude angle of the target to be recognized as a label. Therefore, no attitude angle input is required in the test stage, thus avoiding the problem that the performance of the target recognition model is reduced due to the lack of attitude angle information in the test stage in the related art, and improving the performance of the target recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] 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 for use in the embodiments or the description of the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0061] Figure 1 It is a schematic diagram of the HRRP neural network training method based on auxiliary information in the related art.

[0062] Figure 2 It is a schematic flowchart of the radar HRRP target recognition method based on attitude angle auxiliary information provided by the embodiments of the present invention.

[0063] Figure 3 It is a schematic flowchart of the training process of the target recognition model provided by the embodiments of the present invention.

[0064] Figure 4 It is a schematic diagram of the network structure of the initial target recognition model provided by the embodiments of the present invention.

[0065] Figure 5It is a schematic structural diagram of a radar HRRP target recognition device based on attitude angle auxiliary information provided by an embodiment of the present invention.

[0066] Figure 6 It is a schematic physical structure diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0067] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying 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 based on the embodiments in the present invention without making creative efforts shall fall within the scope of protection of the present invention.

[0068] Figure 1 It is a schematic diagram of a HRRP neural network training method based on auxiliary information in the related art. As Figure 1 shown, the HRRP neural network includes a convolution module 1, a convolution module 2, and a specific task layer connected in sequence. Among them, both the convolution module 1 and the convolution module 2 include a convolution layer, a batch normalization layer, an activation function layer 1, and a max pooling layer connected in sequence. The specific task layer includes a fully connected layer 1, an activation function layer 2, and a fully connected layer 2 connected in sequence. The input of the convolution module 1 includes the sample HRRP data of the sample target and the attitude angle corresponding to the sample HRRP data, and the output of the specific task layer is the category of the sample target. However, in the related art, when training the HRRP neural network, since both the sample HRRP data and the attitude angle corresponding to the sample HRRP data need to be used as input data, the training set and the test set are required to have attitude angles. However, in the real scenario, the echo received by the radar does not contain the corresponding attitude angle. Therefore, in the test stage, the input corresponding to the attitude angle will be set to null, which will result in the missing of the attitude angle information in the test stage, thereby reducing the performance of the finally trained target recognition model.

[0069] Based on this, the present invention proposes a radar HRRP target recognition method based on attitude angle auxiliary information. The radar HRRP data of the target to be recognized is input into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network. Then, the feature information is input into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network. Among them, the target recognition model is trained based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output the predicted category, and the second specific task network is used to output the predicted attitude angle. It can be seen that when training the target recognition model in the present invention, the first specific task network and the second specific task network jointly adopt the output information of the parameter sharing network, and this output information is obtained based on the input sample radar HRRP data, and it is not necessary to input the attitude angle of the target to be recognized. Only the attitude angle of the target to be recognized is used as a label. Therefore, it is not necessary to input the attitude angle of the target to be recognized during the test phase, thus avoiding the problem that the performance of the target recognition model is reduced due to the lack of attitude angle information during the test phase in the related art, and improving the performance of the target recognition model.

[0070] The following combines Figures 2 - 4 to describe the radar HRRP target recognition method based on attitude angle auxiliary information of the present invention. The execution subject of the radar HRRP target recognition method based on attitude angle auxiliary information can be an electronic device such as a terminal, a tablet computer, or a computer, or it can be a radar HRRP target recognition device based on attitude angle auxiliary information provided in the electronic device. The radar HRRP target recognition device based on attitude angle auxiliary information can be implemented by software, hardware, or a combination of both.

[0071] Figure 2 is a schematic flowchart of the radar HRRP target recognition method based on attitude angle auxiliary information provided by an embodiment of the present invention. As Figure 2 shown, the radar HRRP target recognition method based on attitude angle auxiliary information includes the following steps:

[0072] Step 201, obtain the radar HRRP data of the target to be recognized.

[0073] Exemplarily, the acquisition and processing of HRRP data are of great significance for target recognition because HRRP data contains information related to the size of the target, the physical structure of the target, and the distribution of scattering points. In practical applications, HRRP data can be obtained by the method of microwave anechoic chamber measurement.

[0074] Step 202: Input the radar HRRP data into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network.

[0075] The parameter sharing network includes a first convolutional module, a second convolutional module, a third convolutional module, and a first fully connected layer connected in sequence. The first convolutional module, the second convolutional module, and the third convolutional module each include a convolutional layer, a batch normalization layer, a first activation function layer, and a max pooling layer connected in sequence.

[0076] Exemplarily, input the radar HRRP data into the parameter sharing network, and perform feature extraction through the first convolutional module, the second convolutional module, and the third convolutional module in sequence. The output of the first convolutional module can be represented by the following formula (9):

[0077]

[0078] where represents the output of the first convolutional module, represents the radar HRRP data, represents the output of the convolutional layer of the first convolutional module, represents the output of the batch normalization layer of the first convolutional module, represents the activation function.

[0079] After obtaining the output of the third convolutional module, perform feature dimensionality reduction on the output of the third convolutional module through the first fully connected layer with dropout, and finally obtain the feature information output by the first fully connected layer, that is, the feature information output by the parameter sharing network. The above first convolutional module, second convolutional module, third convolutional module, and first fully connected layer together constitute the parameter sharing network, and the shared parameter of the parameter sharing network is .

[0080] Step 203: Input the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network.

[0081] The target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the pose angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted pose angle.

[0082] Exemplarily, when obtaining the feature information output by the parameter-sharing network, input the feature information into the first specific task network of the target recognition model, and output the recognition result of the target category through the first specific task network. Of course, the feature information can also be input into the second specific task network of the target recognition model, and output the pose angle estimation of the target to be recognized through the second specific task network. If the pose angle estimation and category recognition are regarded as two tasks, where, the t network parameters of the th specific task network are t The output of the th specific task network can be represented by the following formula (10):

[0083] (10)

[0084] where, represents the network parameters of the t th specific task network, represents the feature information output by the parameter-sharing network, represents the t bias term of the last fully connected layer in the

[0085] th specific task network. It should be noted that the present invention does not limit the number of convolutional modules included in the parameter-sharing network. For example, it can also be four convolutional modules, etc., which can be specifically set based on requirements.

[0086] The radar HRRP target recognition method based on attitude angle auxiliary information provided by the present invention inputs the radar HRRP data of the target to be recognized into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network, and then inputs the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network. Among them, the target recognition model is obtained by training the initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output the predicted category, and the second specific task network is used to output the predicted attitude angle. It can be seen that when training the target recognition model in the present invention, the first specific task network and the second specific task network jointly adopt the output information of the parameter sharing network, and this output information is obtained based on the input sample radar HRRP data, and it is not necessary to input the attitude angle of the target to be recognized. Only the attitude angle of the target to be recognized is used as an auxiliary information as a label. Therefore, it is not necessary to input the attitude angle of the target to be recognized during the test phase, so as to avoid the problem that the performance of the target recognition model is reduced due to the lack of attitude angle information during the test phase in the related art, and improve the performance of the target recognition model.

[0087] In one embodiment, Figure 3 is a schematic diagram of the training process of the target recognition model provided by the embodiment of the present invention. As Figure 3 shown, the target recognition model is obtained by training in the following manner:

[0088] Step 301, obtain the sample radar HRRP data of the sample target.

[0089] Exemplarily, the sample radar HRRP data of each sample target can be obtained by the method of microwave anechoic chamber measurement.

[0090] Step 302, input the sample radar HRRP data into the parameter sharing network of the initial target recognition model to obtain the sample feature information output by the parameter sharing network.

[0091] Exemplarily, when the sample radar HRRP data of each sample target is obtained, the sample radar HRRP data of each sample target is input into the parameter sharing network of the initial target recognition model, and feature extraction is realized through the first convolution module, the second convolution module, and the third convolution module of the initial target recognition model in sequence, and feature dimensionality reduction is performed through the first fully connected layer with dropout, and finally the sample feature information output by the parameter sharing network of the initial target recognition model is obtained.

[0092] Step 303: Input the sample feature information into the first specific task network and the second specific task network of the initial target recognition model respectively, to obtain the predicted category output by the first specific task network and the predicted pose angle output by the second specific task network.

[0093] Exemplarily, when obtaining the sample feature information output by the parameter sharing network of the initial target recognition model, input the sample feature information into the first specific task network of the initial target recognition model, output the predicted category through the first specific task network, and input the sample feature information into the second specific task network of the initial target recognition model, output the predicted pose angle through the second specific task network.

[0094] Step 304: Determine the total loss function based on the predicted category, the category label of the sample target, the predicted pose angle, and the pose angle label corresponding to the sample target.

[0095] Exemplarily, when obtaining the predicted category and the predicted pose angle of the sample target, construct the total loss function based on the difference between the predicted category of the sample target and the category label of the sample target, and the difference between the predicted pose angle of the sample target and the pose angle label corresponding to the sample target.

[0096] Step 305: Based on the total loss function, iteratively adjust the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network, to obtain the target recognition model.

[0097] Exemplarily, when obtaining the total loss function, the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network can be iteratively adjusted based on the total loss function until the convergence condition is reached, and finally the target recognition model is obtained.

[0098] It should be noted that in each iteration process, the gradient of the shared parameter is the weighted sum of multiple task gradients, which can be specifically represented by the following formula (11):

[0099] (11)

[0100] Where represents the shared parameter with respect to the total loss function gradient, represents the number of tasks, here , represents the th task weight, represents the th task loss function.

[0101] Shared parameters Updated through the gradient descent process: , where represents the learning rate. During the training phase, it is necessary to calculate the recognition task loss function based on the predicted category and the category label, and also calculate the angle estimation task loss function based on the predicted pose angle and the pose angle label. By weighting the recognition task loss function and the angle estimation task loss function, the multi-task joint loss, that is, the total loss function, can be obtained. During the prediction phase, only the radar HRRP data of the target to be recognized needs to be input to obtain the recognition result of the target category and the pose angle estimation of the target to be recognized.

[0102] Figure 4 is a schematic diagram of the network structure of the initial target recognition model provided by the embodiments of the present invention. As Figure 4 shown, the initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. Among them, both the first specific task network and the second specific task network are connected to the parameter sharing network. The parameter sharing network includes a first convolution module, a second convolution module, a third convolution module, and a first fully connected layer connected in sequence. The first convolution module, the second convolution module, and the third convolution module all include a convolution layer, a batch normalization layer, a first activation function layer, and a max pooling layer connected in sequence. The first specific task network and the second specific task network both include a second fully connected layer, a second activation function layer, and a third fully connected layer connected in sequence. The input of the first convolution module is the sample radar HRRP data. The first specific task network is used to output the predicted category, and the second specific task network is used to output the predicted pose angle. Based on the predicted category, the category label of the sample target, the predicted pose angle, and the pose angle label corresponding to the sample target, the total loss function is determined. Based on the total loss function, the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network are iteratively adjusted to obtain the target recognition model. The parameter sharing network adopts a hard parameter sharing architecture. Because the recognition task has a high correlation with the angle estimation task, using the hard parameter sharing architecture can reduce the number of parameters and computational resources without reducing the feature extraction effect.

[0103] It should be noted that the initial target recognition model can also be called an Adaptive Auxiliary Learning Net (AALNet), and the present invention does not limit this.

[0104] In this embodiment, when training the target recognition model, the first specific task network and the second specific task network jointly adopt the output information of the parameter sharing network, and this output information is obtained based on the input sample radar HRRP data. There is no need to input the attitude angle. Only the attitude angle, which is auxiliary information, is used as a label. Therefore, there is no need to input the attitude angle during the test phase either. This can avoid the problem that the performance of the target recognition model decreases due to the lack of attitude angle information during the test phase in the related art, and improve the performance of the trained target recognition model.

[0105] In one embodiment, step 304 determines the total loss function based on the predicted category, the category label of the sample target, the predicted attitude angle, and the attitude angle label corresponding to the sample target. Specifically, it can be implemented in the following manner:

[0106] Based on the predicted category and the category label of the sample target, determine the recognition task loss function; based on the predicted attitude angle and the attitude angle label corresponding to the sample target, determine the angle estimation task loss function; determine the total loss function based on the following formula (1):

[0107] (1)

[0108] Wherein, represents the total loss function, represents the network parameters of the target recognition model, and represent trainable parameters, represents the angle estimation task loss function, represents the recognition task loss function.

[0109] Exemplarily, the present invention relates to two tasks, namely the recognition task and the angle estimation task. The recognition task is the main task, and the recognition task can be regarded as a classification problem. The angle estimation task is an auxiliary task, and the angle estimation task can be regarded as a regression problem. Based on the Gaussian maximum likelihood of homoscedastic uncertainty, the joint objective function under multiple tasks is derived. Assume that the initial target recognition model outputs when the input is the sample radar HRRP data and the network parameters are . The two tasks have the same input. In the case of the outputs of multiple specific task networks, assume that the task scenario satisfies two conditions: is a sufficient statistic, and the labels of the

[0110] (12)

[0111] Among them, represents the probability density function of the output of the probability density function of the output of the first task, represents the probability density function of the output of the

[0112] For tasks of different categories, there are usually different output forms, so they will correspond to different distribution models. In maximum likelihood inference, it is necessary to maximize the likelihood function of the model and derive the weighted loss of multiple tasks considering homoscedastic uncertainty.

[0113] For a regression problem, the output of the regression problem can be modeled as a Gaussian distribution with observational noise, which can be specifically represented by the following formula (13):

[0114] (13)

[0115] Among them, represents the uncertainty parameter, which is also a trainable parameter. As the scale of observational noise, it is fixed in the weight decay of the neural network, so it can be adjusted by the method of maximum likelihood inference , and the maximum likelihood of the objective function is to minimize the negative log-likelihood of the model. Specifically, the above formula (13) can be solved through the following formula (14):

[0116] (14)

[0117] Among them, represents equivalence, represents the label corresponding to the task.

[0118] In addition, usually, the likelihood of the output is adjusted by using the softmax function. Therefore, for a classification problem, the output of the classification problem can be modeled as a Gibbs distribution with a temperature coefficient, which can be specifically represented by the following formula (15):

[0119] (15)

[0120] Among them, the parameter is used to describe the flatness of the distribution, and is usually called the temperature coefficient.

[0121] It can be solved for the above formula (15) through the following formula (16):

[0122] (16)

[0123] in, represents the category label, Represents a category set Any category in , yes An element in a vector.

[0124] In a multi-task scenario with 2 tasks, the two tasks are the recognition task of class prediction of sample targets and the angle estimation task of attitude angle estimation of sample targets. The recognition task is regarded as a classification problem with discrete output, while the angle estimation task is regarded as a regression problem with continuous output. Therefore, continuous output needs to be obtained at the same time. and discrete outputs The joint loss function can be expressed as follows:

[0125] (17)

[0126] in, Represents the joint loss function, that is, the total loss function, represents the network parameters of the target recognition model, and Represents a trainable parameter.

[0127] Furthermore, it is known that two independent tasks are expressed as The mean square error loss and Cross entropy loss At the same time, we can assume the following formula (18) based on experience:

[0128]

[0129] when When it approaches 1, the two are equal. Therefore, the total loss function can be expressed as follows:

[0130] (19)

[0131] in, represents the angle estimation task loss function, Represents the recognition task loss function, which dynamically adjusts the importance of the two tasks during network training.

[0132] Extending formula (19) to the general solution in a multi-task scenario where M outputs are modeled as Gaussian distribution and Gibbs distribution, it can be specifically expressed by the following formula (20):

[0133] (20)

[0134] Among them, K represents the number of regression tasks modeled as Gaussian distribution in the total number of M tasks, and the remaining tasks are all classification tasks modeled as Gibbs distribution, with the number being M - K. Since the loss shown in the above formula (20) will have the phenomenon that the loss function degenerates to 0, modifying the regularization term can meet the condition that the multi-task joint loss is non-negative. Therefore, the above formula (20) is further expressed by the following formula (21):

[0135] (21)

[0136] Further, since the present invention only includes two tasks, the above formula (21) can be expressed by the above formula (1). Through the expression of the total loss function of formula (1), it is easy to know that only by setting the initialization of the parameters, the weights between different tasks can be adaptively and dynamically adjusted during training according to the information amount contained in different tasks and the homoscedastic uncertainty, without the need for manual selection of the weights of the tasks.

[0137] In this embodiment, a method for weighting the loss function when there is a primary-secondary relationship between multiple tasks is designed, and the homoscedastic uncertainty is used to evaluate the importance degree of different tasks, so as to better identify the primary-secondary relationship of multiple tasks and further improve the performance of the target recognition model obtained by training.

[0138] In one embodiment, the above step 301 of obtaining the sample radar HRRP data of the sample target can be specifically implemented by the following method:

[0139] Construct an optimization problem based on the following formula (2):

[0140] (2)

[0141] Among them, the constraint condition is , represents the total number of sample targets, represents the network parameters of the target recognition model, and represent the trainable parameters, represents the sample selection threshold, represents the th control parameter of the sample target, is expressed by the following formula (3):

[0142] (3)

[0143] represents the regularization term, , represents the The multitask joint loss of a sample target,

[0144] , represents the weight of the angle estimation task, represents the angle estimation task loss function corresponding to the th sample target, represents the pose angle label corresponding to the th sample target, represents the predicted pose angle corresponding to the th sample target, represents the network parameters of the second specific task network, represents the recognition task loss function corresponding to the th sample target, represents the classification label corresponding to the th sample target, represents the predicted category corresponding to the th sample target,

[0145] Solving the optimization problem to obtain the network parameters of the target recognition model .

[0146] Exemplarily, in the present invention, the recognition task and the angle estimation task appear as the main task and the auxiliary task respectively. The primary and secondary nature of the tasks makes it unreasonable for the loss function to be adaptively weighted in a simple multitask manner. At the same time, in the early stage of network training, if the weight of the auxiliary task is too large, it will cause the network training direction to deviate, making it easier to fall into a local optimal solution. Therefore, the present invention uses the Self-paced strategy to improve the weighted loss function using uncertainty as the task weight, that is, to improve the total loss function shown in the above formula (1). At the same time, a sample selection method suitable for this scenario is proposed. By the self-paced strategy of the model, the selection order of training samples "from easy to difficult" is realized. In each iteration, simple training samples are preferentially selected and parameters are learned, which can effectively avoid the algorithm from converging to a local optimal state, reduce the training difficulty of the model, and accelerate the fitting speed.

[0147] In the application scenario of the present invention, simple samples and difficult samples are defined as follows: When the category of the sample target is correctly classified and the difference between the predicted pose angle and the pose angle label does not exceed a preset threshold, the sample radar HRRP data of the sample target is defined as a simple sample; when the category of the sample target is misclassified, regardless of the effect of the predicted pose angle, the sample radar HRRP data of the sample target is defined as a difficult sample. The training set is defined as , where, Denote the th training sample, and denote the true label corresponding to the th training sample. Denote the loss function of the input data when the true label is . Among them, denotes the predicted output corresponding to the th training sample, and is the network parameter. Then, the optimization problem shown in the above formula (2) can be obtained. Since the main task contains more information, it should have a greater weight than the auxiliary task. Therefore, is introduced to denote the weight of the auxiliary task, that is, the weight of the angle estimation task. For example, is set to 0.3. For the optimization problem of formula (2), the Alternative Convex Search (ACS) can be used to implement it. Through the iterative bi-convex optimization method, is fixed respectively, and the optimization problem of formula (2) is solved to obtain the network parameter of the target recognition model. When the network parameter is fixed, the global optimal solution of is represented by the above formula (3). By judging whether the recognition task loss function of the main task is greater than the sample selection threshold, it is determined whether the sample radar HRRP data of the sample target is a difficult sample that is difficult to correctly identify. If the sample radar HRRP data is a difficult sample, then is taken as 0, that is, the sample is not included in the training process of the network for the time being. As the network training progresses, the threshold

[0148] is continuously increased by setting an increasing function with the training round cycle, so as to realize the selection of training samples from easy to difficult. In the early stage of training, the selection of simple samples is emphasized, so that the target recognition model is trained in a direction beneficial to the main task. In the middle and late stages of training, difficult samples are gradually introduced to strengthen the fine-tuning of the auxiliary task on the target recognition model and improve the training results.

[0149] In one embodiment, the above-mentioned method for determining the loss function of the angle estimation task based on the predicted pose angle and the pose angle label corresponding to the sample target can be specifically implemented in the following manner:

[0150] Determine the interval width based on the following formula (4):

[0151] (4)

[0152] in, represents the width of the interval, Indicates the maximum angle within the angle space, Indicates the minimum angle within the angle space range, Indicates the total number of pre-set intervals, represents the mapping function;

[0153] The discrete angle corresponding to the predicted attitude angle of the sample target is determined based on the following formula (5):

[0154] (5)

[0155] in, Represents the predicted attitude angle The corresponding discrete angle;

[0156] The distance-penalized metric of the discrete angle is determined based on the following formula (6):

[0157] (6)

[0158] in, represents the metric after distance penalty, Indicates the serial number of the interval in which the predicted attitude angle is located, represents weight;

[0159] The discrete angle is determined based on the following formula (7): Distance penalty error relative to different intervals:

[0160] (7)

[0161] in, Represents the discrete angle Relative to the distance penalty error of different intervals, Indicates the The serial number of the interval;

[0162] The angle estimation task loss function is determined based on the distance penalty error of the discrete angle, the predicted attitude angle, and the attitude angle label corresponding to the sample target.

[0163] Exemplarily, as a non-convex optimization problem, the regression problem makes it difficult to obtain an accurate estimate of unknown variables when the solution space of the model is infinitely continuous, resulting in sub-optimization. At the same time, when estimating variables with periodic or ordinal properties such as attitude angles and discrete heights, the mean squared error loss cannot well represent the gap between the predicted variable and the actual variable. To solve this problem, ordinal regression is usually introduced to model the task instead of the regression problem. The present invention proposes a cosine-weighted ordinal regression-based azimuth prediction loss function to replace the original mean squared error loss as the loss function of the auxiliary task in multi-task learning. Since it is difficult for the regression problem to accurately predict the target variable under the condition of an infinite solution space, the present invention considers starting from the perspective of quantization, transforming the infinite space of the regression problem solution into a finite discrete space for estimating the probability density. Essentially, the regression problem is transformed into a special classification problem through discretization means, which is generally divided into the following five steps: identity mapping, interval partitioning, distance penalty metric, probability distribution, and cosine weighting.

[0164] First, it is determined that the range of the angle space to be predicted is from 0 degrees to 360 degrees. Then, the maximum angle within the angle space range is 360 degrees, and the minimum angle within the angle space range is 0 degrees. The angle is mapped to a discrete angle by means of identity mapping. Specifically, the interval width can be calculated through the above formula (4), and the interval width is substituted into the above formula (5) to calculate the discrete angle corresponding to the predicted attitude angle of the sample target. Here the value of can be 36. Suppose the predicted attitude angle of the sample target is 121 degrees, and

[0165] is 10 degrees, then = 12.1. Then, when is obtained, the angle space range can be partitioned into intervals, which can be expressed as where represents the first interval, with a value of 0, indicating the serial number of the first interval, represents the th interval, with a value of

[0166] and indicating the serial number of the th interval. Next, a distance penalty metric is used to measure the relationship between the discrete angle The distance penalty error with respect to different intervals is further used to determine the angle estimation task loss function based on the distance penalty error of the discrete angle, the predicted pose angle, and the pose angle label corresponding to the sample target.

[0167] In this embodiment, the predicted pose angle of the sample target is discretized to obtain the discrete angle corresponding to the predicted pose angle of the sample target, and the metric after distance penalty of the discrete angle is calculated. Then, the distance penalty error of the discrete angle with respect to different intervals is calculated based on the metric after distance penalty of the discrete angle. Finally, the angle estimation task loss function is determined based on the distance penalty error of the discrete angle, the predicted pose angle, and the pose angle label corresponding to the sample target, so as to solve the problem of large estimation error caused by the infinite continuous solution space of the traditional regression problem.

[0168] In one embodiment, the above-mentioned determination of the angle estimation task loss function based on the distance penalty error of the discrete angle, the predicted pose angle, and the pose angle label corresponding to the sample target can be specifically implemented in the following manner:

[0169] Perform a mapping process on the distance penalty error of the discrete angle to obtain the probability density of the discrete angle in different intervals;

[0170] Determine the angle estimation task loss function based on the following formula (8):

[0171] (8)

[0172] where, represents the angle estimation task loss function, represents the total number of sample targets, represents the th probability density of the discrete angle of the th sample target in different intervals, represents the vector of the pose angle label of the th sample target, represents the cosine weight of the th sample target, represents the th pose angle label of the

[0173] th sample target. For example, when obtaining the distance penalty error of the discrete angle, the distance penalty error of the discrete angle can be mapped through the softmax process to obtain the probability density of the discrete angle in different intervals. Since the pose angle is a variable with inherent periodicity, the present invention uses the cosine weighting method to process the probability density of the discrete angle interval to obtain a loss function that conforms to the angle periodicity. Specifically, the cosine weight of the th sample target, when the pose angle label Cosine value and predicted attitude angle When the difference between the cosine value and the cosine value of the predicted attitude angle is closer, The smaller it is, that is, the smaller the loss value corresponding to the sample target. By controlling the cosine weight to control the loss function of different sample targets, the adaptation to the periodic problem of angle estimation is realized. The loss function of the finally obtained cosine-weighted ordered regression is as shown in the above formula (8). Taking the loss function of the cosine-weighted ordered regression as the loss function of the angle estimation task, the form of the loss function of the cosine-weighted ordered regression can be simply regarded as the weighted cross-entropy loss, which is used to replace the original mean square error loss function to calculate the loss of angle estimation.

[0174] In this embodiment, the loss function of the cosine-weighted ordered regression is used to replace the mean square error loss function originally used for angle estimation, which solves the problem of the infinite solution space of angle estimation and the periodic problem of angle estimation.

[0175] In summary, the present invention can, in the case of missing attitude angles during the test process, only use the sample radar HRRP data, and use the attitude angle and category as labels to complete the training of the target recognition model, so as to realize the improvement of the performance of the main task by using auxiliary information; at the same time, the sample selection method of the self-paced strategy is used to design and implement the selection of the sample radar HRRP data in the training set during the training process, and the simple samples are preferentially trained through the training idea of "from easy to difficult", restricting the target recognition model to fit towards the main task direction, and reducing the possibility of falling into local optimal points; the loss function of the angle estimation task is improved. The commonly used mean square error loss cannot well describe the error between different attitude angles because the attitude angle has the characteristic of periodicity. The present invention proposes cosine-weighted ordered regression, which improves the estimation effect in the form of discretized intervals, and at the same time, cosine weighting for each interval can balance the problems brought by periodicity. The AALNet network of the present invention is compared with other single-task networks, and the effect is significantly better than other methods. Table 1 is the performance comparison between the AALNet network of the present invention and other networks. It can be seen from Table 1 that the accuracy of the AALNet network trained based on Dataset 1 is greater than the accuracy of other networks, the accuracy of the AALNet network trained based on Dataset 2 is greater than the accuracy of other networks, and when the measured data is input into the AALNet network, the prediction accuracy of the AALNet network is also greater than the accuracy of other networks.

[0176] Table 1

[0177]

[0178] Among them, AGC-LSTM is the Adaptive Gaussian Classifier-Long Short-Term Memory, Deep RF is the Deep Random Forest, MIConvGRU is the Multi-Input Convolutional Gated Recurrent Unit Neural Network, and CNN-Bi-RNN is the Convolutional Neural Network-Bidirectional Recurrent Neural Network.

[0179] The radar HRRP target recognition device based on attitude angle auxiliary information provided by the present invention will be described below. The radar HRRP target recognition device based on attitude angle auxiliary information described below can be correspondingly referred to the radar HRRP target recognition method based on attitude angle auxiliary information described above.

[0180] Figure 5 It is a schematic structural diagram of the radar HRRP target recognition device based on attitude angle auxiliary information provided by an embodiment of the present invention. As Figure 5 shown, the radar HRRP target recognition device 500 based on attitude angle auxiliary information includes an acquisition unit 501, an extraction unit 502, and an identification unit 503; where:

[0181] The acquisition unit 501 is configured to acquire radar HRRP data of a target to be recognized;

[0182] The extraction unit 502 is configured to input the radar HRRP data into a parameter sharing network of a target recognition model to obtain feature information output by the parameter sharing network;

[0183] The identification unit 503 is configured to input the feature information into a first specific task network of the target recognition model to obtain an identification result of a target category output by the first specific task network;

[0184] Among them, the target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. Both the first specific task network and the second specific task network are connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted attitude angle.

[0185] The radar HRRP target recognition device based on attitude angle auxiliary information provided by the present invention inputs the radar HRRP data of the target to be recognized into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network, and then inputs the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network. Among them, the target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. Both the first specific task network and the second specific task network are connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted attitude angle. It can be seen that when training the target recognition model in the present invention, the first specific task network and the second specific task network jointly adopt the output information of the parameter sharing network, and this output information is obtained based on the input sample radar HRRP data, and it is not necessary to input the attitude angle of the target to be recognized, but only use the attitude angle of the target to be recognized as a label. Therefore, it is not necessary to input the attitude angle of the target to be recognized during the test phase, thus avoiding the problem that the performance of the target recognition model is reduced due to the lack of attitude angle information during the test phase in the related art, and improving the performance of the target recognition model.

[0186] Based on any of the above embodiments, the target recognition model is trained in the following manner:

[0187] Obtain the sample radar HRRP data of the sample target;

[0188] Input the sample radar HRRP data into the parameter sharing network of the initial target recognition model to obtain the sample feature information output by the parameter sharing network;

[0189] Input the sample feature information into the first specific task network and the second specific task network of the initial target recognition model respectively to obtain the predicted category output by the first specific task network and the predicted attitude angle output by the second specific task network;

[0190] Determine a total loss function based on the predicted category, the category label of the sample target, the predicted pose angle, and the pose angle label corresponding to the sample target;

[0191] Based on the total loss function, iteratively adjust the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network to obtain the target recognition model.

[0192] Based on any of the above embodiments, the determining the total loss function based on the predicted category, the category label of the sample target, the predicted pose angle, and the pose angle label corresponding to the sample target includes:

[0193] Determine an identification task loss function based on the predicted category and the category label of the sample target;

[0194] Determine an angle estimation task loss function based on the predicted pose angle and the pose angle label corresponding to the sample target;

[0195] Determine the total loss function based on the following formula (1):

[0196] (1)

[0197] Wherein, represents the total loss function, represents the network parameters of the target recognition model, and represent trainable parameters, represents the angle estimation task loss function, represents the identification task loss function.

[0198] Based on any of the above embodiments, the obtaining the sample radar HRRP data of the sample target includes:

[0199] Construct an optimization problem based on the following formula (2):

[0200] (2)

[0201] Wherein, the constraint condition is , represents the total number of sample targets, represents the network parameters of the target recognition model, and represent trainable parameters, represents a sample selection threshold, represents the th control parameter of the sample target, It is represented by the following formula (3):

[0202] (3)

[0203] represents the regularization term, , represents the multitask joint loss of the th sample target,

[0204] , represents the weight of the angle estimation task, represents the th sample target corresponding angle estimation task loss function, represents the th sample target corresponding pose angle label, represents the th sample target corresponding predicted pose angle, represents the network parameters of the second specific task network, represents the th sample target corresponding recognition task loss function, represents the th sample target corresponding classification label, represents the th sample target corresponding predicted category, represents the network parameters of the first specific task network;

[0205] Solve the optimization problem to obtain the network parameters of the target recognition model .

[0206] Based on any of the above embodiments, determining the angle estimation task loss function based on the predicted pose angle and the pose angle label corresponding to the sample target includes:

[0207] Determine the interval width based on the following formula (4):

[0208] (4)

[0209] wherein, represents the interval width, represents the maximum angle within the angle space range, represents the minimum angle within the angle space range, represents the total number of preset intervals, represents the mapping function;

[0210] Determine the discrete angle corresponding to the predicted pose angle of the sample target based on the following formula (5):

[0211] (5)

[0212] Among them, represents the discrete angle corresponding to the predicted attitude angle corresponding discrete angle;

[0213] Determine the metric after distance penalty for the discrete angle based on the following formula (6):

[0214] (6)

[0215] Among them, represents the metric after distance penalty, represents the serial number of the interval where the predicted attitude angle is located, represents the weight;

[0216] Determine the discrete angle relative to the distance penalty error of different intervals based on the following formula (7):

[0217] (7)

[0218] Among them, represents the discrete angle relative to the distance penalty error of different intervals, represents the serial number of the k-th interval;

[0219] Determine the loss function of the angle estimation task based on the distance penalty error of the discrete angle, the predicted attitude angle, and the attitude angle label corresponding to the sample target.

[0220] Based on any of the above embodiments, determining the loss function of the angle estimation task based on the distance penalty error of the discrete angle, the predicted attitude angle, and the attitude angle label corresponding to the sample target includes:

[0221] Perform mapping processing on the distance penalty error of the discrete angle to obtain the probability density of the discrete angle in different intervals;

[0222] Determine the loss function of the angle estimation task based on the following formula (8):

[0223] (8)

[0224] Among them, represents the loss function of the angle estimation task, represents the total number of sample targets, represents the probability density of the discrete angle of the n-th sample target in different intervals, represents the A vector of pose angle labels for a sample target, indicating the cosine weight of the th sample target, and indicating the pose angle label of the

[0225] Based on any of the above embodiments, the parameter sharing network includes a first convolutional module, a second convolutional module, a third convolutional module, and a first fully connected layer connected in sequence. The first convolutional module, the second convolutional module, and the third convolutional module each include a convolutional layer, a batch normalization layer, a first activation function layer, and a max pooling layer connected in sequence. The first specific task network and the second specific task network each include a second fully connected layer, a second activation function layer, and a third fully connected layer connected in sequence.

[0226] Figure 6 FIG. is a schematic physical structure diagram of an electronic device provided by an embodiment of the present invention. As Figure 6 shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640. Among them, the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call logic instructions in the memory 630 to execute a radar HRRP target recognition method based on pose angle auxiliary information. The method includes: obtaining radar HRRP data of a target to be recognized;

[0227] inputting the radar HRRP data into the parameter sharing network of the target recognition model to obtain feature information output by the parameter sharing network;

[0228] inputting the feature information into the first specific task network of the target recognition model to obtain an identification result of the target category output by the first specific task network;

[0229] wherein, the target recognition model is obtained by training an initial target recognition model based on sample radar HRRP data of sample targets, category labels of the sample targets, and pose angle labels corresponding to the sample targets. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted pose angle.

[0230] In addition, when the logical instructions in the above-mentioned memory 630 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0231] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the radar HRRP target recognition method based on attitude angle auxiliary information provided by the above-mentioned various methods. The method includes: obtaining radar HRRP data of a target to be recognized;

[0232] Inputting the radar HRRP data into a parameter sharing network of a target recognition model to obtain feature information output by the parameter sharing network;

[0233] Inputting the feature information into a first specific task network of the target recognition model to obtain an identification result of the target category output by the first specific task network;

[0234] Among them, the target recognition model is obtained by training an initial target recognition model based on sample radar HRRP data of sample targets, category labels of the sample targets, and attitude angle labels corresponding to the sample targets. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted attitude angle.

[0235] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to execute the radar HRRP target recognition method based on attitude angle auxiliary information provided by the above-mentioned various methods. The method includes: obtaining radar HRRP data of a target to be recognized;

[0236] Input the radar HRRP data into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network;

[0237] Input the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network;

[0238] Among them, the target recognition model is obtained by training an initial target recognition model based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted attitude angle.

[0239] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0240] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing 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.

[0241] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A radar HRRP target recognition method based on attitude angle auxiliary information, characterized in that Including: Obtain the radar HRRP data of the target to be recognized; Input the radar HRRP data into the parameter sharing network of the target recognition model to obtain the feature information output by the parameter sharing network; Input the feature information into the first specific task network of the target recognition model to obtain the recognition result of the target category output by the first specific task network; Wherein, the target recognition model is trained based on the sample radar HRRP data of the sample target, the category label of the sample target, and the attitude angle label corresponding to the sample target. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. Both the first specific task network and the second specific task network are connected to the parameter sharing network. The first specific task network is used to output the predicted category, and the second specific task network is used to output the predicted attitude angle.

2. The radar HRRP target recognition method based on attitude angle auxiliary information according to claim 1, wherein The target recognition model is trained based on the following method: Obtain the sample radar HRRP data of the sample target; Input the sample radar HRRP data into the parameter sharing network of the initial target recognition model to obtain the sample feature information output by the parameter sharing network; Input the sample feature information into the first specific task network and the second specific task network of the initial target recognition model respectively to obtain the predicted category output by the first specific task network and the predicted attitude angle output by the second specific task network; Determine the total loss function based on the predicted category, the category label of the sample target, the predicted attitude angle, and the attitude angle label corresponding to the sample target; Based on the total loss function, iteratively adjust the shared parameters of the parameter sharing network, the network parameters of the first specific task network, and the network parameters of the second specific task network to obtain the target recognition model.

3. The radar HRRP target recognition method based on attitude angle auxiliary information according to claim 2, wherein The determining the total loss function based on the predicted category, the category label of the sample target, the predicted attitude angle, and the attitude angle label corresponding to the sample target includes: Determine the recognition task loss function based on the predicted category and the category label of the sample target; Determine the angle estimation task loss function based on the predicted attitude angle and the attitude angle label corresponding to the sample target; Determine the total loss function based on the following formula (1): （1） Among them, represents the total loss function, represents the network parameters of the target recognition model, and represents the trainable parameters, represents the angle estimation task loss function, represents the recognition task loss function.

4. The radar HRRP target recognition method based on attitude angle auxiliary information according to claim 2, characterized in that The method further includes: Construct an optimization problem based on the following formula (2); （2） Among them, the constraint condition is , represents the total number of sample targets, represents the network parameters of the target recognition model, and represent trainable parameters, represents the sample selection threshold, represents the th control parameter of the sample target, which is expressed by the following formula (3): （3） denotes the regularization term, , denotes the multi-task joint loss of the th sample target, , represents the weight of the angle estimation task, denotes the loss function of the angle estimation task corresponding to the th sample target, denotes the pose angle label corresponding to the th sample target, denotes the predicted pose angle corresponding to the th sample target, represents the network parameters of the second specific task network, denotes the th sample target's recognition task loss function, denotes the classification label corresponding to the th sample target, denotes the predicted category corresponding to the th sample target, represents the network parameters of the first specific task network; Solve the optimization problem to obtain the network parameters of the target recognition model .

5. The radar HRRP target recognition method based on attitude angle auxiliary information according to claim 3, characterized in that, The determining the angle estimation task loss function based on the predicted attitude angle and the attitude angle label corresponding to the sample target includes: Determine the interval width based on the following formula (4); （4） Among them, represents the interval width, represents the maximum angle within the angular space range, represents the minimum angle within the angular space range, represents the total number of preset intervals, represents the mapping function; Determine the discrete angle corresponding to the predicted attitude angle of the sample target based on the following formula (5); （5） Among them, represents the discrete angle corresponding to the predicted attitude angle ; Determine the distance-penalized metric of the discrete angle based on the following formula (6); （6） Among them, represents the metric after distance penalty, represents the serial number of the interval where the predicted attitude angle is located, represents the weight; Determine the discrete angle based on the following formula (7) Distance penalty error with respect to different intervals （7） Among them, represents the discrete angle with respect to the distance penalty error of different intervals, represents the serial number of the k-th interval; Determine the angle estimation task loss function based on the distance-penalized error of the discrete angle, the predicted attitude angle, and the attitude angle label corresponding to the sample target.

6. The radar HRRP target recognition method based on attitude angle auxiliary information according to claim 5, wherein, Determining the angle estimation task loss function based on the distance penalty error of the discrete angles, the predicted pose angle, and the pose angle label corresponding to the sample target includes: Performing mapping processing on the distance penalty error of the discrete angles to obtain the probability density of the discrete angles in different intervals; Determining the angle estimation task loss function based on the following formula (8): （8） Among them, represents the loss function of the angle estimation task, represents the total number of sample targets, represents the probability density of the discrete angle of the th sample target in different intervals, represents the vector of the attitude angle label of the th sample target, represents the cosine weight of the , represents the th sample target's attitude angle label, represents the th sample target's predicted attitude angle.

7. The radar HRRP target recognition method based on attitude angle auxiliary information according to any one of claims 1-6, characterized in that, The parameter sharing network includes a first convolution module, a second convolution module, a third convolution module, and a first fully connected layer connected in sequence. The first convolution module, the second convolution module, and the third convolution module each include a convolution layer, a batch normalization layer, a first activation function layer, and a max pooling layer connected in sequence. The first specific task network and the second specific task network each include a second fully connected layer, a second activation function layer, and a third fully connected layer connected in sequence.

8. A radar HRRP target recognition device based on attitude angle auxiliary information, characterized in that Including: An acquisition unit for acquiring radar HRRP data of a target to be recognized; An extraction unit for inputting the radar HRRP data into the parameter sharing network of the target recognition model to obtain feature information output by the parameter sharing network; A recognition unit for inputting the feature information into the first specific task network of the target recognition model to obtain a recognition result of the target category output by the first specific task network; Wherein, the target recognition model is obtained by training an initial target recognition model based on sample radar HRRP data of sample targets, category labels of the sample targets, and pose angle labels corresponding to the sample targets. The initial target recognition model includes a parameter sharing network, a first specific task network, and a second specific task network. The first specific task network and the second specific task network are both connected to the parameter sharing network. The first specific task network is used to output a predicted category, and the second specific task network is used to output a predicted pose angle.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the radar HRRP target recognition method based on pose angle auxiliary information according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the radar HRRP target recognition method based on pose angle auxiliary information according to any one of claims 1 to 7.

Images