Radar signal classification method, device, electronic equipment, medium and program product

Abstract

The present invention provides a radar signal classification method, apparatus, electronic device, medium, and program product, relating to the field of radar application technology. The method comprises: determining multiple initial hyperparameter combinations; calculating the fitness value of a classification model corresponding to each initial hyperparameter combination using a radar signal sample dataset to determine an initial optimal hyperparameter combination; updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination based on the initial optimal hyperparameter combination, adaptive weights, and the hyperparameter combination to be updated; iteratively updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination; determining model parameters of a classification model based on the initial optimal hyperparameter combination obtained when a preset iteration condition is met to obtain an optimized classification model; and classifying the radar signal to be classified using the optimized classification model to obtain a classification result of the radar signal. The present invention improves the accuracy of the classification results.

Description

Technical Field

[0001] The present invention relates to the field of radar application technology, and more specifically, to a radar signal classification method, device, electronic equipment, medium, and program product. Background Art

[0002] Radar systems detect and identify aircraft by emitting electromagnetic waves and receiving reflected echoes from targets. With the development of modern aviation technology, radar signals have multimodal, time-varying, and nonlinear characteristics, making traditional radar signal classification methods less accurate.

[0003] Aircraft radar echo signals are a core data source for aviation safety monitoring, and their classification results directly influence target identification, threat assessment, and flight decision-making. Radar echo signals are susceptible to interference in complex electromagnetic environments, such as strong noise, multipath interference, and weather clutter. When a target is in motion, such as when its posture changes, the radar echo signal exhibits nonlinear, time-varying characteristics.

[0004] Traditional radar signal classification methods rely on features obtained through manual intervention. For example, random forests classify radar echo signals based on the spectral energy obtained through manual intervention. However, they have difficulty capturing the spatiotemporal coupling characteristics of radar echo signals, resulting in reduced accuracy in radar echo signal classification.

[0005] In addition, radar signal classification methods based on feature engineering use methods such as Fourier transform and wavelet analysis to extract the time and frequency domain characteristics of radar signals, and then use classifiers such as support vector machines and decision trees for identification. However, this method relies on features obtained through manual intervention and has poor adaptability to complex radar signals, resulting in reduced accuracy in radar echo signal classification.

[0006] Therefore, there is an urgent need for a method that can improve the accuracy of radar echo signal classification. Summary of the Invention

[0007] In view of this, the present invention provides a radar signal classification method, device, electronic device, medium and program product.

[0008] One aspect of the present invention provides a radar signal classification method, comprising: determining a plurality of initial hyperparameter combinations based on respective search ranges of a plurality of hyperparameters; each of the initial hyperparameter combinations including the same number of hyperparameters; different initial hyperparameter combinations having different values ​​of at least a portion of the hyperparameters; each initial hyperparameter combination corresponding to a classification model; the classification model including a convolutional neural network and a long short-term memory network; calculating the fitness value of the classification model corresponding to each initial hyperparameter combination using a radar signal sample data set, and taking the initial hyperparameter combination corresponding to the maximum fitness value as the initial optimal hyperparameter combination; and performing an update on the plurality of initial hyperparameter combinations and the initial optimal hyperparameter combination based on the initial optimal hyperparameter combination, an adaptive weight, and a hyperparameter combination to be updated selected from the plurality of initial hyperparameter combinations. The adaptive weight is determined by the maximum value of the distance between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the boundary of the search range; the new multiple hyperparameter combinations are used as the multiple initial hyperparameter combinations, the new initial optimal hyperparameter combination is used as the initial optimal hyperparameter combination, and the operation for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination is iteratively performed until a preset iteration condition is met; the model parameters of the classification model are determined based on the new initial optimal hyperparameter combination obtained when the preset iteration condition is met, and an optimized classification model is obtained; the radar signal to be classified is classified using the optimized classification model to obtain a classification result of the radar signal.

[0009] Another aspect of the present invention provides a radar signal classification device, comprising: an initial parameter determination module, for determining a plurality of initial hyperparameter combinations based on respective search ranges of a plurality of hyperparameters; each of the aforementioned initial hyperparameter combinations includes a plurality of hyperparameters of the same number; the values ​​of at least a portion of the hyperparameters in different initial hyperparameter combinations are different; each initial hyperparameter combination corresponds to a classification model; the classification model includes a convolutional neural network and a long short-term memory network; an optimal parameter determination module, for calculating the fitness value of the classification model corresponding to each initial hyperparameter combination using a radar signal sample data set, and taking the initial hyperparameter combination corresponding to the maximum fitness value as the initial optimal hyperparameter combination; a parameter combination updating module, for updating the aforementioned multiple initial hyperparameter combinations and the aforementioned initial optimal hyperparameter combination based on the aforementioned initial optimal hyperparameter combination, adaptive weights, and a hyperparameter combination to be updated selected from the aforementioned multiple initial hyperparameter combinations; The combination is updated to obtain new multiple hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the above-mentioned adaptive weight is determined by the above-mentioned initial optimal hyperparameter combination, the above-mentioned hyperparameter combination to be updated, and the maximum value of the distance between the above-mentioned initial optimal hyperparameter combination and the boundary of the above-mentioned search range; the optimal parameter iteration module is used to use the new multiple hyperparameter combinations as the above-mentioned multiple initial hyperparameter combinations and the new initial optimal hyperparameter combination as the above-mentioned initial optimal hyperparameter combination, and iteratively perform operations for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination until a preset iteration condition is reached; the classification model determination module is used to determine the model parameters of the above-mentioned classification model according to the new initial optimal hyperparameter combination obtained when the preset iteration condition is reached, and obtain an optimized classification model; the radar signal classification module is used to classify the radar signal to be classified using the above-mentioned optimized classification model to obtain a classification result of the radar signal.

[0010] Another aspect of the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0011] Another aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0012] Another aspect of the present invention provides a computer program product, which includes computer executable instructions. When the instructions are executed, they are used to implement the above method.

[0013] According to an embodiment of the present invention, the process of obtaining model parameters of the classification model undergoes multiple iterative calculations. During each iterative calculation, the fitness value of the classification model is calculated, a global optimum is searched from multiple initial hyperparameter combinations, and the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination are dynamically updated using adaptive weights to avoid falling into local optimal solutions. This allows model parameters that result in a high classification accuracy for the classification model, thereby obtaining an optimized classification model. Since the multiple iterative calculations are performed automatically without manual intervention, the efficiency of determining the model parameters of the classification model is improved. Since the convolutional neural network included in the classification model can extract the spatial characteristics of the radar signal, and the long short-term memory network included in the classification model can capture the dynamic temporal characteristics of the radar signal, the classification model can classify the radar signal to be classified based on the spatiotemporal characteristics of the radar signal to be classified, thereby obtaining a more accurate classification result. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0015] Figure 1 An exemplary system architecture to which the radar signal classification method and apparatus of the present invention can be applied is shown.

[0016] Figure 2 A flowchart of a radar signal classification method according to an embodiment of the present invention is shown.

[0017] Figure 3 A structural diagram of an optimized classification model according to an embodiment of the present invention is shown.

[0018] Figure 4 The flowchart of the radar signal classification method according to another embodiment of the present invention is shown.

[0019] Figure 5 A block diagram of a radar signal classification device according to an embodiment of the present invention is shown.

[0020] Figure 6 A block diagram of an electronic device suitable for implementing a radar signal classification method according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0021] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concept of the present invention.

[0022] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the presence of the features, steps, operations and / or components, but do not exclude the presence or addition of one or more other features, steps, operations or components.

[0023] In the embodiments of the present invention, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of all data involved (including, but not limited to, user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken to prevent unauthorized access to user personal information data and maintain the security of user personal information and network security.

[0024] In the embodiment of the present invention, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0025] Deep learning-based aircraft radar signal classification methods include those based on convolutional neural networks (CNNs) and long-short-term memory networks (LSTMs). These methods automatically learn multi-level features directly from raw radar echo signals without manual intervention and perform well in complex electromagnetic environments. CNN-based classification methods directly process radar time-frequency images and automatically extract spatial features through convolution. However, due to CNNs' limited ability to capture temporal information, they struggle to process dynamic changes in the signal. LSTM-based classification methods excel at processing time-series signals, using gating mechanisms to memorize long-term dependencies in the data and capture dynamic temporal characteristics in radar signals. However, LSTMs perform poorly when processing time-series signals with very long time steps. Since radar signal time-series data can contain thousands of time steps, the gradient signal can vanish or explode during backpropagation due to the long path length, making it difficult for LSTMs to effectively capture dependencies in radar signals over such long time series. In addition, the training of radar signal classification methods based on deep learning requires a lot of resources and time, and is prone to falling into local optimality. Its hyperparameters need to be fine-tuned. For example, the size of the hidden layer needs to be fine-tuned to avoid falling into local optimality. The hyperparameters of the deep learning network rely on manual tuning, which is inefficient.

[0026] An embodiment of the present invention provides a radar signal classification method, comprising: determining a plurality of initial hyperparameter combinations according to respective search ranges of a plurality of hyperparameters; each initial hyperparameter combination comprises a plurality of hyperparameters of the same number; at least a portion of the hyperparameters in different initial hyperparameter combinations have different values; each initial hyperparameter combination corresponds to a classification model; the classification model comprises a convolutional neural network and a long short-term memory network; calculating the fitness value of the classification model corresponding to each initial hyperparameter combination using a radar signal sample data set, and taking the initial hyperparameter combination corresponding to the maximum fitness value as the initial optimal hyperparameter combination; and performing an update on the plurality of initial hyperparameter combinations and the initial optimal hyperparameter combination according to the initial optimal hyperparameter combination, an adaptive weight, and a hyperparameter combination to be updated selected from the plurality of initial hyperparameter combinations. The parameter combination is updated to obtain new multiple hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the adaptive weight is determined by the maximum value of the distance between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the boundary of the search range; the new multiple hyperparameter combinations are used as multiple initial hyperparameter combinations, the new initial optimal hyperparameter combination is used as the initial optimal hyperparameter combination, and the operation for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination is iteratively performed until a preset iteration condition is met; based on the new initial optimal hyperparameter combination obtained when the preset iteration condition is met, model parameters of the classification model are determined to obtain an optimized classification model; and the radar signal to be classified is classified using the optimized classification model to obtain a classification result of the radar signal.

[0027] According to an embodiment of the present invention, the process of obtaining the model parameters of the classification model undergoes multiple iterative calculations. During each iterative calculation, the fitness value of the classification model is calculated, and a global optimum is searched from multiple initial hyperparameter combinations. The multiple initial hyperparameter combinations are dynamically updated using adaptive weights to avoid falling into local optimal solutions, thereby obtaining model parameters that increase the classification accuracy of the classification model, and thus obtaining an optimized classification model. Since the multiple iterative calculations are performed automatically without human intervention, the efficiency of determining the model parameters of the classification model is improved. Since the convolutional neural network included in the classification model can extract the spatial characteristics of the radar signal, and the long short-term memory network included in the classification model can capture the dynamic time series characteristics in the radar signal, the classification model can classify the radar signal to be classified according to the spatiotemporal characteristics of the radar signal to be classified, thereby obtaining a more accurate classification result.

[0028] Figure 1 The following shows an exemplary system architecture to which the radar signal classification method and apparatus of the present invention can be applied. Figure 1 The examples shown are merely examples of system architectures to which the embodiments of the present invention may be applied, to help those skilled in the art understand the technical content of the present invention, but do not mean that the embodiments of the present invention cannot be used in other devices, systems, environments or scenarios.

[0029] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a laser radar 101, a first terminal device 1021, a second terminal device 1022, a third terminal device 1023, a network (not shown in the figure), and a server 103. The network is used as a medium for providing a communication link between the first terminal device 1021, the second terminal device 1022, the third terminal device 1023 and the laser radar 101 or the server 103. The network may include various connection types, such as wired and / or wireless communication links, etc.

[0030] The user can use the first terminal device 1021, the second terminal device 1022, and the third terminal device 1023 to interact with the server 103 or the laser radar 101 through the network to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 1021, the second terminal device 1022, and the third terminal device 1023, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software (for example only).

[0031] The first terminal device 1021 , the second terminal device 1022 , and the third terminal device 1023 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0032] The server 103 may be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 1021, the second terminal device 1022, and the third terminal device 1023. The backend management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0033] It should be noted that the radar signal classification method provided in the embodiments of the present invention can generally be executed by the server 103. Accordingly, the radar signal classification device provided in the embodiments of the present invention can generally be located in the server 103. The radar signal classification method provided in the embodiments of the present invention can also be executed by a server or server cluster that is different from the server 103 and that is capable of communicating with the first terminal device 1021, the second terminal device 1022, the third terminal device 1023, and / or the server 103. Accordingly, the radar signal classification device provided in the embodiments of the present invention can also be located in a server or server cluster that is different from the server 103 and that is capable of communicating with the first terminal device 1021, the second terminal device 1022, the third terminal device 1023, and / or the server 103. Alternatively, the radar signal classification method provided in the embodiments of the present invention can also be executed by the first terminal device 1021, the second terminal device 1022, or the third terminal device 1032, or by another terminal device different from the first terminal device 1021, the second terminal device 1022, or the third terminal device 1023. Correspondingly, the radar signal classification device provided in the embodiment of the present invention can also be set in the first terminal device 1021, the second terminal device 1022 or the third terminal device 1023, or in other terminal devices different from the first terminal device 1021, the second terminal device 1022 or the third terminal device 1023.

[0034] For example, the radar signal to be classified may be originally stored in any one of the first terminal device 1021, the second terminal device 1022, or the third terminal device 1023 (e.g., the first terminal device 1021, but not limited thereto), or stored on an external storage device and imported into the first terminal device 1021. The first terminal device 1021 may then locally execute the radar signal classification method provided in the embodiment of the present invention, or send the radar signal to be classified to another terminal device, server, or server cluster, which then executes the radar signal classification method provided in the embodiment of the present invention on the other terminal device, server, or server cluster that receives the raw power spectrum data.

[0035] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0036] Figure 2 FIG. 1 is a flow chart showing a method for classifying radar signals according to an embodiment of the present invention. Figure 2 As shown, the method includes operations S201 to S206.

[0037] In operation S201, multiple initial hyperparameter combinations are determined based on the respective search ranges of the multiple hyperparameters; each initial hyperparameter combination includes the same number of hyperparameters; the values ​​of at least some of the hyperparameters in different initial hyperparameter combinations are different; each initial hyperparameter combination corresponds to a classification model; the classification model includes a convolutional neural network and a long short-term memory network.

[0038] According to an embodiment of the present invention, taking multiple hyperparameters as A, B, and C, each initial hyperparameter combination includes A, B, and C. Each initial hyperparameter combination can be expressed as (value of A, value of B, value of C). Taking multiple initial hyperparameter combinations as an example, the first initial hyperparameter combination can be expressed as (1, 1, 2), and the second initial hyperparameter combination can be expressed as (1, 1, 1).

[0039] According to an embodiment of the present invention, the multiple hyperparameters of the classification model include at least one of the following: the number of convolutional neural network (CNN) convolutional layers, the convolution kernel size of each convolutional layer, the number of LSTM (Long Short-Term Memory Network) layers, the number of units in each LSTM layer, the learning rate of the training hyperparameter, and the batch size. The discrete values ​​of the hyperparameters, which are discrete data, are converted into continuous values ​​by transcoding, and the number of initial hyperparameter combinations included in the multiple initial hyperparameter combinations is set.

[0040] According to an embodiment of the present invention, formula (1) is applied to determine multiple initial hyperparameter combinations of the classification model.

[0041] (1)

[0042] Among them, lb is the lower boundary of the hyperparameter search space, ub is the upper boundary of the hyperparameter search space, Rand is a random vector with elements between [0, 1], and the dimension of Rand is the number of hyperparameters. is the initial hyperparameter combination, i is an integer in the range [1, n], and n is the number of initial hyperparameter combinations. Each initial hyperparameter combination includes the same number of hyperparameters, but each hyperparameter has a different meaning.

[0043] According to an embodiment of the present invention, the classification model includes a convolutional neural network, a long short-term memory network and an attention layer; the long short-term memory network includes an input layer, a long short-term memory network layer and an output layer; the convolutional neural network is embedded between the input layer and the long short-term memory network layer; and the attention layer is embedded between the long short-term memory network layer and the output layer.

[0044] In operation S202 , the radar signal sample data set is used to calculate the fitness value of the classification model corresponding to each initial hyperparameter combination, and the initial hyperparameter combination corresponding to the maximum fitness value is used as the initial optimal hyperparameter combination.

[0045] According to an embodiment of the present invention, a radar signal sample dataset can be obtained through simulation or experimentation. For example, simulation is used to generate radar echo signals for different aircraft types. Interference noise, such as thermal noise, clutter, and electromagnetic interference, is added to the generated radar echo signals. Radar echo signals from different aircraft are distinguished using labels to construct an aircraft radar echo dataset. The aircraft radar echo dataset is then divided into training, validation, and test sets in a 7:2:1 ratio. The training, validation, and test sets of the aircraft radar echo dataset are preprocessed, with the preprocessed training set serving as the radar signal sample dataset, the preprocessed validation set serving as the validation set for the trained classification model, and the preprocessed test set serving as the test set for the validated classification model. Aircraft types can be categorized by purpose, for example, civil aircraft, drones, and other aircraft; by flight mode and lift generation principle, for example, helicopters and fixed-wing aircraft; and by radar cross-section.

[0046] According to an embodiment of the present invention, an aircraft radar echo dataset includes multiple class labels and multiple echo sequence sample signals corresponding to each class label. The class label is the aircraft type code. For example, the class label for other aircraft is 1, the class label for civil aircraft is 2, and the class label for drones is 3. The echo sequence sample signals are radar echo signals of the aircraft corresponding to the class label.

[0047] The preprocessing process for the aircraft radar echo dataset includes: preprocessing each echo sequence sample signal in the training set, validation set, and test set to obtain multiple preprocessed echo sequence sample signals corresponding to each class label; segmenting each preprocessed echo sequence sample signal in the training set, validation set, and test set to obtain multiple subsequence samples of the same length; applying a normalization method to normalize each subsequence sample in the training set to obtain multiple normalized echo sequences corresponding to each class label in the training set. The multiple class labels and corresponding normalized echo sequences in the training set are used as the preprocessed radar echo dataset. Based on the normalization method, the mean and variance of the normalized training set are applied to normalize each subsequence sample in the validation set and test set to obtain multiple normalized echo sequences corresponding to each class label in the validation set and test set, thus forming the preprocessed validation set and test set.

[0048] Specifically, each echo sequence sample signal is preprocessed to obtain multiple preprocessed echo sequence sample signals corresponding to each category label, including: supplementing missing values ​​of each echo sequence sample signal; deleting invalid data and outliers of each echo sequence sample signal after the supplementation operation; filtering and denoising each echo sequence sample signal after the deletion operation to obtain multiple preprocessed echo sequence sample signals.

[0049] According to an embodiment of the present invention, a sliding window method is used to segment each preprocessed echo sequence sample signal to obtain multiple subsequence samples of the same length. Each subsequence sample is normalized using the Z-Score normalization method to obtain multiple normalized echo sequences corresponding to each class label. The formula for the Z-Score normalization method is shown in Formula (2).

[0050] (2)

[0051] Among them, x is a subsequence sample, is the mean of multiple subsequence samples of the training set, σ is the variance of multiple subsequence samples of the training set, is the normalized echo sequence.

[0052] According to an embodiment of the present invention, the fitness value can be obtained by calculating the evaluation index of the classification model. For example, the fitness value is set as a loss function or an accuracy rate. The fitness value is obtained by inputting each initial hyperparameter combination to be evaluated into the classification model, applying the training set for training, and calculating the evaluation index of the trained classification model. Specifically, when the fitness value is set as the loss function, the loss function of the trained classification model is calculated as the fitness value. When the fitness value is set as the accuracy rate, the accuracy rate of the trained classification model is calculated as the fitness value.

[0053] According to an embodiment of the present invention, after obtaining the fitness value corresponding to each initial hyperparameter combination to be evaluated, multiple fitness values ​​are compared, the largest fitness value is selected from the multiple fitness values, and the initial hyperparameter combination corresponding to the largest fitness value is used as the initial optimal hyperparameter combination corresponding to the current number of iterations.

[0054] In operation S203, multiple initial hyperparameter combinations and the initial optimal hyperparameter combination are updated according to the initial optimal hyperparameter combination, the adaptive weight, and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations to obtain new multiple hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the adaptive weight is determined by the maximum value of the distance between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the initial optimal hyperparameter combination and the boundary of the search range.

[0055] According to an embodiment of the present invention, a plurality of initial hyperparameter combinations and an initial optimal hyperparameter combination are obtained; the search space boundary of each hyperparameter is determined; the distance from the initial optimal hyperparameter combination to each boundary is calculated, and the maximum value is taken as the reference distance; for each hyperparameter combination to be updated, its distance from the initial optimal hyperparameter combination is calculated; an adaptive weight is determined based on the distance and the reference distance, and the adaptive weight can be obtained by S-type function mapping, and the longer the distance, the smaller the weight; the initial optimal hyperparameter combination and the hyperparameter combination to be updated are weighted using the adaptive weight to obtain an updated hyperparameter combination; the centroid of all updated hyperparameter combinations is used as a temporary optimal solution, and is moved closer to the initial optimal solution according to a preset ratio to obtain a new initial optimal hyperparameter combination.

[0056] In operation S204, the new multiple hyperparameter combinations are used as multiple initial hyperparameter combinations, the new initial optimal hyperparameter combination is used as the initial optimal hyperparameter combination, and the operation for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination is iteratively performed until the preset iteration condition is reached.

[0057] According to an embodiment of the present invention, the preset iteration condition is to iteratively execute operation S203 until a preset number of iterations is reached, with each execution of operation S203 being counted as one iteration until the preset iteration condition is satisfied. The preset iteration condition can be a preset number of iterations, which is a preset maximum number of iterations and can be further set based on needs or actual circumstances.

[0058] In operation S205 , model parameters of the classification model are determined according to the new initial optimal hyperparameter combination obtained when the preset iteration condition is met, thereby obtaining an optimized classification model.

[0059] According to an embodiment of the present invention, taking the multiple hyperparameters of the classification model including the number of convolutional layers of the convolutional neural network, the convolution kernel size of each convolutional layer, the number of LSTM layers of the long short-term memory network, the number of units of each LSTM layer, the learning rate of the training hyperparameters, and the batch size as an example, for example, the optimal hyperparameter combination obtained when the preset iteration condition is reached is [3, 7, 7, 15, 15, 7, 7, 1, 61, 0.0007202583283431076, 64]. In the figure, 3 is the number of convolutional layers, 7 is the width of the convolution kernel in the first convolutional layer, 7 is the height of the convolution kernel in the first convolutional layer, 15 is the width of the convolution kernel in the second convolutional layer, 15 is the height of the convolution kernel in the second convolutional layer, 7 is the width of the convolution kernel in the third convolutional layer, 7 is the height of the convolution kernel in the third convolutional layer, 1 is the number of LSTM layers, 61 is the number of units in the LSTM layer, 0.0007202583283431076 is the learning rate, and 64 is the batch size.

[0060] Figure 3 A structural diagram of an optimized classification model according to an embodiment of the present invention is shown.

[0061] like Figure 3As shown, the optimized classification model includes an input layer 310, a CNN network 320, a long short-term memory network 330, a self-attention layer 340, a linear layer 350, a fourth activation function layer 360, and an output layer 370, which are connected in sequence. The CNN network 320 includes a first convolutional layer 321, a normalization layer 322, a first activation function layer 323, a second convolutional layer 324, a dropout layer 325, a second activation function layer 326, a third convolutional layer 327, a max pooling layer 328, and a third activation function layer 329, which are connected in sequence. The input layer 310 is used to receive input data; the convolutional layer is used to extract local features in the spatial domain of the input data, capturing static spatial features such as echo intensity distribution and edge mutations; the pooling layer compresses the dimension of the feature vector; the long short-term memory network 330 extracts the time-dependent characteristics of the data output by the CNN network 320 through a gating mechanism; the activation function introduces a nonlinear component; and the output layer integrates and maps the extracted features into the final classification result. A random dropout mechanism is used between layers of the model to prevent overfitting.

[0062] The optimizer of the classification model is configured as Adam, the activation function is LeakyReLU, the training batch size is 64, the learning rate is 0.0007202583283431076, the maximum number of iterations is 100, the number of initial hyperparameter combinations is set to 20, the number of units in the LSTM layer is set to the range of [32,512], the number of LSTM layers is set to [1,3], the batch size is set to [32,64,128], and one of the three values ​​is selected. The dropout rate of the dropout layer is set to the range of [0.1,0.5], and the learning rate is set to [10 -5 ,10 -2 ], the numerical values ​​are continuous, and the optimal hyperparameter combination table shown in Table 1 is obtained.

[0063] The new optimal hyperparameter combination table shown in Table 1

[0064]

[0065] Among them, the number of channels of the input layer is 1; the size of the convolution kernel of the first convolution layer is 7×7, the stride is 2, and the number of output channels is 64; the number of channels of the normalization layer is 64, the size of the convolution kernel of the second convolution layer is 15×15, the stride is 1, and the number of output channels is 128; the size of the convolution kernel of the third convolution layer is 7×7, the stride is 1, and the number of output channels is 128; the window size of the maximum pooling layer is 2×2, and the stride is 2; the long short-term memory network includes a long short-term memory layer with 61 units; the number of channels of the linear layer is 128, and the activation functions applied in the first activation function layer, the second activation function layer, the third activation function layer, and the fourth activation function layer are all LeakyReLU activation functions; the number of channels of the output layer is 3.

[0066] In operation S206 , the radar signal to be classified is classified using the optimized classification model to obtain a classification result of the radar signal.

[0067] According to an embodiment of the present invention, hyperparameters are parameters that need to be manually set before training the optimized classification model. Hyperparameters can be structural parameters of the classification model, such as the number of convolutional layers and the size of the convolution kernel of each convolutional layer. By determining the hyperparameters, the structure of the optimized classification model is obtained. Hyperparameters can also include learning rate, batch size, number of training rounds, regularization parameters, optimizer-related parameters, etc. The process of training the optimized classification model is the process of determining the weights and biases of each neuron in the classification model. It can also be the optimization or fine-tuning of certain hyperparameters in the optimized classification model. For example, the learning rate can be fine-tuned by training the optimized classification model through the training set.

[0068] According to an embodiment of the present invention, after obtaining an optimized classification model, the optimized classification model is trained using a preprocessed training set to obtain a trained optimized classification model. The trained classification model is then verified using a preprocessed validation set, and the verified classification model is then tested using a preprocessed test set. When the trained classification model is verified and meets preset verification conditions, and the verified classification model is tested and meets preset test conditions, the trained classification model is used to classify a radar signal to be classified, obtaining a classification result for the radar signal.

[0069] In the case that the trained classification model does not meet the preset verification conditions, the preprocessed training set is applied to continue training the trained classification model until the trained classification model meets the preset verification conditions.

[0070] When training the optimized classification model, the normalized echo sequence of the training set is used as input, and the corresponding class label is used as output. When the optimized classification model is trained a preset number of times or the evaluation indicators meet the preset requirements, the trained classification model is obtained. For example, the accuracy rate reaches above 90%, and the validation set accuracy rate of the validation set used to verify the trained classification model reaches above 90%.

[0071] Applications such as Figure 3 The trained classification model and LSTM network model shown in FIG1 are used to classify the same radar signal to be classified, and the comparison results shown in Table 2 are obtained.

[0072] Table 2 Comparison of classification results

[0073]

[0074] As shown in Table 2, the trained classification model has a training set accuracy of 99.12, a validation set accuracy of 95.84, a test set accuracy of 95.77, a precision of 95.45, a recall rate of 95.01, and an F1-Score of 95.22, all of which are better than the LSTM network model.

[0075] According to an embodiment of the present invention, by combining CNN with LSTM, a CNN-LSTM model is obtained, which can improve the disadvantage of LSTM that it is difficult to process long time series, and can integrate the spatiotemporal characteristics of the aircraft radar echo sequence, thereby improving the CNN-LSTM model's ability to recognize complex echo signals, thereby improving the ability to classify aircraft radar echo signals. The Black Hawk optimization algorithm is used to optimize the hyperparameters of the CNN-LSTM model, which can quickly locate the optimal hyperparameter combination and avoid falling into local optimality and repeated training processes. Through the synergistic effect of the Black Hawk optimization algorithm and the CNN-LSTM model, the dual improvement of feature extraction and model optimization in the aircraft radar echo signal classification task is achieved, providing more reliable technical support for aircraft radar target recognition in complex scenarios.

[0076] According to an embodiment of the present invention, based on the initial optimal hyperparameter combination, the adaptive weight, and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations, multiple initial hyperparameter combinations and the initial optimal hyperparameter combination are updated to obtain new multiple hyperparameter combinations and a new initial optimal hyperparameter combination, including operations S301~S306.

[0077] In operation S301, multiple initial hyperparameter combinations are updated according to the initial optimal hyperparameter combination, the adaptive weight, the target hyperparameter combination selected from the multiple initial hyperparameter combinations, and the hyperparameter combination to be updated to obtain multiple hyperparameter combinations updated for the first time.

[0078] According to an embodiment of the present invention, in the process of obtaining the first updated multiple hyperparameter combinations, the parameter search strategy of the Black Eagle Optimizer (BEO) can be used to guide the search for the global optimum by simulating the hunting behavior of the Black Eagle, thereby obtaining the first updated multiple hyperparameter combinations.

[0079] In operation S302, the radar signal sample data set is used to calculate the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the first time, and the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination are combined to form a fitness value set. The hyperparameter combination corresponding to the largest fitness value in the fitness value set is used as the initial optimal hyperparameter combination to obtain the initial optimal hyperparameter combination after the first update.

[0080] According to an embodiment of the present invention, the method in operation S202 is applied to calculate the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the first time, select the maximum fitness value from the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the first time, compare the maximum fitness value with the fitness value of the classification model corresponding to the initial optimal hyperparameter combination, select the larger fitness value between the two, and use the hyperparameter combination corresponding to the larger fitness value as the initial optimal hyperparameter combination after the first update.

[0081] In operation S303, the multiple hyperparameter combinations updated for the first time are updated according to the initial optimal hyperparameter combination after the first update, the rotation matrix with the same dimension as the initial optimal hyperparameter combination after the first update, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the first time, to obtain multiple hyperparameter combinations updated for the second time.

[0082] According to an embodiment of the present invention, in the process of obtaining multiple hyperparameter combinations for the second update, based on the parameter search strategy of the Black Eagle Optimizer (BEO), by simulating the hovering behavior of the Black Eagle, a rotation matrix is ​​applied to refine the search area, thereby obtaining multiple hyperparameter combinations for the second update.

[0083] In operation S304, the radar signal sample data set is used to calculate the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the second time, and the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination after the first update are combined to form a fitness value set. The hyperparameter combination corresponding to the largest fitness value in the fitness value set is used as the initial optimal hyperparameter combination, and the initial optimal hyperparameter combination after the second update is obtained.

[0084] According to an embodiment of the present invention, the method in operation S202 is applied to calculate the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the second time, select the maximum fitness value from the fitness values ​​of the classification models corresponding to the multiple hyperparameter combinations updated for the second time, compare the maximum fitness value with the fitness value of the classification model corresponding to the initial optimal hyperparameter combination after the first update, select the larger fitness value between the two, and use the hyperparameter combination corresponding to the larger fitness value as the initial optimal hyperparameter combination after the second update.

[0085] In operation S305, the multiple hyperparameter combinations updated for the second time are updated according to the initial optimal hyperparameter combination after the second update, the current number of iterations, the preset number of iterations, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the second time to obtain multiple new hyperparameter combinations.

[0086] According to an embodiment of the present invention, in the process of determining a new optimal hyperparameter combination, the parameter search strategy of the Black Eagle Optimizer (BEO) can be used to simulate the capture behavior of the Black Eagle and dynamically scale the factors. and , gradually converge to the global optimal solution, thus obtaining a new optimal hyperparameter combination.

[0087] In operation S306, the radar signal sample data set is used to calculate the fitness values ​​of the classification models corresponding to each of the new multiple hyperparameter combinations, and the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination after the second update are combined to form a fitness value set. The hyperparameter combination corresponding to the largest fitness value in the fitness value set is used as the initial optimal hyperparameter combination to obtain a new initial optimal hyperparameter combination.

[0088] According to an embodiment of the present invention, the method in operation S202 is applied to calculate the fitness values ​​of the classification models corresponding to the new multiple hyperparameter combinations, select the maximum fitness value from the fitness values ​​of the classification models corresponding to the new multiple hyperparameter combinations, compare the maximum fitness value with the fitness value of the classification model corresponding to the initial optimal hyperparameter combination after the second update, select the larger fitness value between the two, and use the hyperparameter combination corresponding to the larger fitness value as the new initial optimal hyperparameter combination.

[0089] The multiple initial hyperparameter combinations are updated based on the initial optimal hyperparameter combination, the adaptive weight, and the target hyperparameter combination selected from the multiple initial hyperparameter combinations and the hyperparameter combination to be updated, thereby obtaining multiple hyperparameter combinations for a first update. This includes: selecting the target hyperparameter combination and the hyperparameter combination to be updated from the multiple initial hyperparameter combinations. The adaptive weight is determined based on the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the maximum value of the distance between the initial optimal hyperparameter combination and the boundary of the search range. The hyperparameter combination to be updated is updated based on the initial optimal hyperparameter combination, the adaptive weight, the target hyperparameter combination, and the hyperparameter combination to be updated, thereby determining an updated first hyperparameter combination. The updated first hyperparameter combination is applied to replace the hyperparameter combination to be updated from the multiple initial hyperparameter combinations to determine multiple replaced hyperparameter combinations. The replaced multiple hyperparameter combinations are used as the multiple initial hyperparameter combinations, a target hyperparameter combination is selected from the multiple initial hyperparameter combinations, and a target updated hyperparameter combination is selected from the multiple initial hyperparameter combinations other than the updated first hyperparameter combination. Taking the target update hyperparameter combination as the hyperparameter combination to be updated, iteratively perform operations for determining the adaptive weight, determining the first hyperparameter combination after update, determining the multiple hyperparameter combinations after replacement, and selecting the target hyperparameter combination and the target update hyperparameter combination until the first preset iteration condition is met, and the multiple hyperparameter combinations after replacement obtained in the current iteration round when the first preset iteration condition is met are used as the multiple hyperparameter combinations for the first update.

[0090] According to an embodiment of the present invention, multiple initial hyperparameters are updated in an iterative manner. In a first iteration, a hyperparameter combination is selected from the multiple initial hyperparameter combinations as the hyperparameter combination to be updated. The hyperparameter combination to be updated can be determined in a random manner. Then, a target hyperparameter combination is randomly selected from the multiple initial hyperparameter combinations, and the target hyperparameter combination is used as the update direction of the hyperparameter combination to be updated. The target hyperparameter combination and the hyperparameter combination to be updated are not the same initial hyperparameter combination.

[0091] The adaptive weight can be determined using formula (3).

[0092] (3)

[0093] in, is the adaptive weight of the i-th hyperparameter combination to be updated, X best is the initial optimal hyperparameter combination under the current number of iterations, is the i-th hyperparameter combination to be updated, is the distance between the i-th hyperparameter combination to be updated and the initial optimal hyperparameter combination, and D is the initial optimal hyperparameter combination The maximum distance to the search boundary. There are multiple hyperparameters in the calculation, each of which has its own search range. The farthest distance between each hyperparameter and its respective search range is obtained, and the maximum value is selected from multiple farthest distances and used as the value of D.

[0094] Apply formula (4) to determine the updated first hyperparameter combination.

[0095] (4)

[0096] Among them, X inew is the first hyperparameter combination after the update of the i-th hyperparameter combination to be updated, r1 is a random number between 0 and 1, X ik is the target hyperparameter combination.

[0097] For each iteration, the hyperparameter combination selected to be updated is the initial hyperparameter combination that has not been updated. The target hyperparameter combination can be a hyperparameter combination that has not been updated among multiple initial hyperparameter combinations, or an updated hyperparameter combination.

[0098] For example, if multiple initial hyperparameter combinations are [X1, X2, X3], select X2 as the hyperparameter combination to be updated. , select X1 as the target hyperparameter combination, apply formula (2) and formula (3) to get the first hyperparameter combination X 1new , the multiple hyperparameter combinations after replacement are [X1, X 1new , X3]. The replaced multiple hyperparameters are combined into [X1,X 1new ,X3] as multiple initial hyperparameter combinations, from [X1,X 1new ,X3], select X1 or X 1new Or X3 is the target hyperparameter combination, select X1 or X3 as the hyperparameter combination to be updated, take the target hyperparameter combination as X3, and the hyperparameter combination to be updated as X1 as an example, continue to perform the second iteration, and the multiple hyperparameter combinations after replacement are obtained as [X 2new ,X 1new ,X3], until the first preset iteration condition is met.

[0099] According to an embodiment of the present invention, the first preset iteration condition may be a first preset number of iterations, where the first preset number of iterations is the number of initial hyperparameter combinations included in the multiple initial hyperparameter combinations.

[0100] When the first preset iteration condition is the first preset number of iterations, that is, X1, X2, and X3 in the multiple initial hyperparameter combinations [X1, X2, X3] are all updated, corresponding to three iterations, the preset number of iterations is set to 3.

[0101] According to an embodiment of the present invention, multiple hyperparameter combinations updated for the first time are updated according to the initial optimal hyperparameter combination after the first update, the rotation matrix with the same dimension as the initial optimal hyperparameter combination after the first update, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the first time to obtain multiple hyperparameter combinations updated for the second time, including: updating multiple hyperparameter combinations updated for the first time according to the initial optimal hyperparameter combination after the first update, the rotation matrix with the same dimension as the initial optimal hyperparameter combination, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the first time to obtain multiple hyperparameter combinations updated for the second time, including: selecting the hyperparameter combination to be updated from the multiple hyperparameter combinations updated for the first time to obtain a first target updated hyperparameter combination; updating the first target updated hyperparameter combination according to the rotation matrix, the first target updated hyperparameter combination and the initial optimal hyperparameter combination after the first update A new hyperparameter combination is updated to determine the updated first target hyperparameter combination; the updated first target hyperparameter combination is applied to replace the first target update hyperparameter combination in the multiple hyperparameter combinations updated for the second time, and multiple first hyperparameter combinations are determined; the multiple first hyperparameter combinations are used as multiple initial hyperparameter combinations, and the hyperparameter combination to be updated is selected from the multiple initial hyperparameter combinations except the updated first target hyperparameter combination in the multiple initial hyperparameter combinations to obtain a new first target update hyperparameter combination; the new first target update hyperparameter combination is used as the first target update hyperparameter combination, and the operations of determining the updated first target hyperparameter combination, determining multiple first hyperparameter combinations, and selecting the new first target update hyperparameter combination are iteratively performed until the second preset iteration condition is met, and the multiple first hyperparameter combinations obtained in the current iteration round when the second preset iteration condition is met are used as the multiple hyperparameter combinations updated for the second time.

[0102] According to an embodiment of the present invention, a first target updated hyperparameter combination is determined from multiple hyperparameter combinations updated for the first time by a random selection method or a preset order. For example, the multiple hyperparameter combinations updated for the first time are [X 2new ,X 1new ,X 3new ], then when updating in a random manner, X can be updated in the first iteration 1new Update, in the second iteration, X 3new Update, in the third iteration, 2new Update; when updating in the forward order according to the preset order, X is updated in the order of iteration. 2new、X 1new and X 3new to update.

[0103] The dimension of the rotation matrix is ​​equal to the number of hyperparameters in the initial hyperparameter combination. For example, the rotation matrix can be in the form of formula (5).

[0104] (5)

[0105] In formula (5), m is the rotation matrix and a is a random angle between [0, 2π].

[0106] Apply formula (6) to determine the first target hyperparameter combination after update.

[0107] (6)

[0108] in, Update the hyperparameter combination for the i-th first target; The updated first target hyperparameter combination corresponding to the i-th first target update hyperparameter combination, is the initial optimal hyperparameter combination after the first update. j has no specific physical meaning and is only used as a distinguishing mark.

[0109] Formulas (5) and (6) are applied to update each of the multiple hyperparameter combinations updated for the first time. During this update process, the rotation matrix m remains unchanged after the value of a is selected.

[0110] According to an embodiment of the present invention, the second preset iteration condition may be a second preset number of iterations, which is the number of initial hyperparameter combinations included in the multiple initial hyperparameter combinations.

[0111] According to an embodiment of the present invention, according to the initial optimal hyperparameter combination after the second update, the current number of iterations, the preset number of iterations and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations of the second update, the multiple hyperparameter combinations of the second update are updated to obtain new multiple hyperparameter combinations, including: according to the initial optimal hyperparameter combination after the second update, the current number of iterations, the preset number of iterations and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations of the second update, the multiple hyperparameter combinations of the second update are updated to obtain new multiple hyperparameter combinations, including: obtaining a second target update hyperparameter combination from the hyperparameter combination to be updated selected from the multiple hyperparameter combinations of the second update; determining a third target update hyperparameter combination according to the second target update hyperparameter combination, the initial optimal hyperparameter combination after the second update, the current number of iterations and the preset number of iterations; determining a third target update hyperparameter combination according to the third target update hyperparameter combination, the preset number of iterations, the current number of iterations and the initial optimal hyperparameter combination after the second update, determine the fourth target update hyperparameter combination; apply the fourth target update hyperparameter combination to replace the second target update hyperparameter combination in the multiple hyperparameter combinations updated for the second time, and determine multiple second hyperparameter combinations; use the multiple second hyperparameter combinations as the multiple initial hyperparameter combinations, select the hyperparameter combination to be updated from the multiple initial hyperparameter combinations except the fourth target update hyperparameter combination, and obtain a new fourth target update hyperparameter combination; use the new fourth target update hyperparameter combination as the second target update hyperparameter combination, iteratively perform the operations of determining the third target update hyperparameter combination, determining the fourth target update hyperparameter combination, determining multiple second hyperparameter combinations, and selecting the new fourth target update hyperparameter combination until the third preset iteration condition is met, and use the multiple second hyperparameter combinations obtained in the current iteration round when the third preset iteration condition is met as the multiple hyperparameter combinations for the third update.

[0112] According to an embodiment of the present invention, a second target updated hyperparameter combination is determined from the multiple hyperparameter combinations updated for the second time by random selection or a preset order. j 1new , X j 2new , X j 3new ] as an example, then when updating in a random manner, X can be updated in the first iteration. j 1new Update, in the second iteration, X j 3new Update, in the third iteration, j 2new Update; when updating in the forward order according to the preset order, X is updated in the order of iteration.j 1new 、X j 2new and X j 3new Apply formula (7), formula (8), formula (9) and formula (10) to determine the third target update hyperparameter combination.

[0113] (7)

[0114] (8)

[0115] (9)

[0116] (10)

[0117] in, is the initial optimal hyperparameter combination after the second update, In the iterative process of obtaining new multiple hyperparameter combinations, the exploration factor that decays with iteration at the tth iteration is used, and T is the total number of iterations to obtain new multiple hyperparameter combinations; Update the hyperparameter combination for the third objective corresponding to the i-th second objective hyperparameter combination; Update the hyperparameter combination for the i-th second objective. k has no specific physical meaning and is only used as a distinguishing mark. s0 is a random perturbation term that obeys the standard normal distribution. In order to obtain the exploration factor that increases with iteration in the iterative process of the new multiple hyperparameter combinations at the t+1th iteration, Update the hyperparameter combination for the fourth objective corresponding to the i-th second objective update hyperparameter combination.

[0118] According to an embodiment of the present invention, the third preset iteration condition may be a third preset number of iterations, where the third preset number of iterations is the number of initial hyperparameter combinations included in the plurality of initial hyperparameter combinations. t is the current iteration number in the preset number of iterations, and T is the preset number of iterations. Each time operations S301-S306 are performed, t is incremented by 1, for a total of T iterations.

[0119] Figure 4 FIG. 5 shows a flow chart of a radar signal classification method according to another embodiment of the present invention. Figure 4 As shown, the method includes operations S401 to S418.

[0120] In operation S401, a sequence of aircraft radar echoes is generated by simulation, which belongs to the data acquisition part.

[0121] In operation S402 , a data set is constructed. Specifically, a radar signal sample data set is constructed based on the aircraft radar echo sequence.

[0122] In operation S403 , data is cleaned.

[0123] In operation S404, noise reduction processing is performed.

[0124] In operation S405 , the sliding window is segmented.

[0125] In operation S406 , normalization is performed.

[0126] Operations S402 to S406 belong to the data preprocessing part.

[0127] In operation S407 , a hyperparameter search space is defined.

[0128] In operation S408 , the Black Hawk population position is initialized.

[0129] In operation S409 , a fitness function is designed.

[0130] In operation S410 , the population is iteratively optimized by simulating black hawk behavior.

[0131] In operation S411, the population position is updated.

[0132] In operation S412, it is determined whether the maximum number of iterations has been reached or the fitness has converged. If the maximum number of iterations has been reached or the fitness has converged, operation S413 is performed; otherwise, operation S410 is performed.

[0133] In operation S413 , the optimal hyperparameter combination is output.

[0134] Operations S407 to S413 are the steps of searching for the optimal parameter combination.

[0135] In operation S414 , the optimized hyperparameter combination is imported into the classification model.

[0136] In operation S415 , an optimized classification model is obtained.

[0137] In operation S416, the training set is used to train the optimized classification model, the validation set is used to validate the trained classification model, and the test set is used to test the validated classification model. In operation S416, the training set, validation set, and test set are all pre-processed.

[0138] In operation S417, the best training model is output.

[0139] Operations S414 to S417 belong to the model training part.

[0140] In operation S418, a classification level test of the aircraft radar echo signal is performed, which belongs to the model prediction part.

[0141] According to an embodiment of the present invention, the CNN-LSTM model, as a classification model, utilizes the CNN convolutional layer to extract local features of the input data, efficiently extracting key features and reducing the redundant information that the LSTM needs to process. This can overcome the LSTM's difficulty in processing very long time series. Furthermore, the global search capability of the Black Hawk optimization algorithm is utilized to optimize the hyperparameters of the CNN-LSTM model, automatically and quickly locating the optimal hyperparameter combination that is more suitable for the current dataset, ensuring optimal performance of the optimized classification model.

[0142] Figure 5 FIG. 1 shows a block diagram of a radar signal classification device according to an embodiment of the present invention. Figure 5 As shown, the radar signal classification device 500 includes an initial parameter determination module 510 , an optimal parameter determination module 520 , a parameter combination update module 530 , an optimal parameter iteration module 540 , a classification model determination module 550 and a radar signal classification module 560 .

[0143] The initial parameter determination module 510 is used to determine multiple initial hyperparameter combinations based on the search ranges of the multiple hyperparameters; each initial hyperparameter combination includes the same number of hyperparameters; the values ​​of at least some of the hyperparameters in different initial hyperparameter combinations are different; each initial hyperparameter combination corresponds to a classification model; the classification model includes a convolutional neural network and a long short-term memory network.

[0144] The optimal parameter determination module 520 is used to calculate the fitness value of the classification model corresponding to each initial hyperparameter combination using the radar signal sample data set, and use the initial hyperparameter combination corresponding to the maximum fitness value as the initial optimal hyperparameter combination.

[0145] The parameter combination updating module 530 is used to update multiple initial hyperparameter combinations and the initial optimal hyperparameter combination according to the initial optimal hyperparameter combination, the adaptive weight, and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations, to obtain multiple new hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the adaptive weight is determined by the maximum value of the distance between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the initial optimal hyperparameter combination and the boundary of the search range.

[0146] The optimal parameter iteration module 540 is used to use the new multiple hyperparameter combinations as multiple initial hyperparameter combinations and the new initial optimal hyperparameter combination as the initial optimal hyperparameter combination, and iteratively perform operations for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination until a preset iteration condition is met.

[0147] The classification model determination module 550 is used to determine the model parameters of the classification model according to the new initial optimal hyperparameter combination obtained when the preset iteration conditions are met, so as to obtain the optimized classification model.

[0148] The radar signal classification module 560 is used to classify the radar signal to be classified using the optimized classification model to obtain a classification result of the radar signal.

[0149] Any number of the modules, submodules, units, and subunits according to embodiments of the present invention, or at least part of the functionality of any number of these units, can be implemented in a single module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be split into multiple modules for implementation. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware using any other reasonable method of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as a computer program module that, when executed, can perform the corresponding functionality.

[0150] For example, any number of the initial parameter determination module 510, the optimal parameter determination module 520, the parameter combination update module 530, the optimal parameter iteration module 540, the classification model determination module 550, and the radar signal classification module 560 may be combined into a single module / unit / sub-unit for implementation, or any one of these modules / units / sub-units may be split into multiple modules / units / sub-units. Alternatively, at least part of the functionality of one or more of these modules / units / sub-units may be combined with at least part of the functionality of other modules / units / sub-units and implemented in a single module / unit / sub-unit. According to an embodiment of the present invention, at least one of the initial parameter determination module 510, the optimal parameter determination module 520, the parameter combination update module 530, the optimal parameter iteration module 540, the classification model determination module 550, and the radar signal classification module 560 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or may be implemented in hardware or firmware by any other reasonable means of circuit integration or packaging, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of these. Alternatively, at least one of the initial parameter determination module 510, the optimal parameter determination module 520, the parameter combination update module 530, the optimal parameter iteration module 540, the classification model determination module 550, and the radar signal classification module 560 may be at least partially implemented as a computer program module, which, when executed, may perform the corresponding function.

[0151] It should be noted that the data processing system part in the embodiment of the present invention corresponds to the data processing method part in the embodiment of the present invention. The description of the data processing system part specifically refers to the data processing method part and will not be repeated here.

[0152] Figure 6 A block diagram of an electronic device suitable for implementing a radar signal classification method according to an embodiment of the present invention is shown. Figure 6 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0153] like Figure 6As shown, an electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage unit 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0154] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes the programs in ROM 602 and / or RAM 603 to perform various operations according to the method flow of the embodiment of the present invention. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute the programs stored in one or more memories to perform various operations according to the method flow of the embodiment of the present invention.

[0155] According to an embodiment of the present invention, electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to bus 604. Electronic device 600 may also include one or more of the following components connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 608 including a hard disk; and a communication section 609 including a network interface card such as a LAN card or modem. Communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. Removable media 611, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 610 as needed, so that computer programs read from the removable media can be installed into storage section 608 as needed.

[0156] According to an embodiment of the present invention, the method flow according to an embodiment of the present invention can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above-mentioned functions defined in the system of the embodiment of the present invention are executed. According to an embodiment of the present invention, the system, device, apparatus, module, unit, etc. described above can be implemented by a computer program module.

[0157] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0158] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0159] For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 602 and / or the RAM 603 described above and / or one or more memories other than the ROM 602 and the RAM 603 .

[0160] An embodiment of the present invention also includes a computer program product, which includes a computer program containing program code for executing the method provided by the embodiment of the present invention. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the radar signal classification method provided by the embodiment of the present invention.

[0161] When the computer program is executed by the processor 601, the above functions defined in the system / device of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0162] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0163] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0164] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, as well as the combination of boxes in the block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or may be implemented using a combination of dedicated hardware and computer instructions. It will be understood by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention may be combined and / or coupled in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0165] The above describes embodiments of the present invention. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A radar signal classification method, characterized in that: The classification method includes: Determining a plurality of initial hyperparameter combinations based on respective search ranges of the plurality of hyperparameters; each of the initial hyperparameter combinations includes the same number of hyperparameters; different initial hyperparameter combinations have different values ​​of at least some of the hyperparameters; each initial hyperparameter combination corresponds to a classification model; the classification model includes a convolutional neural network and a long short-term memory network; The radar signal sample dataset is used to calculate the fitness value of the classification model corresponding to each initial hyperparameter combination, and the initial hyperparameter combination corresponding to the largest fitness value is taken as the initial optimal hyperparameter combination; updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination according to the initial optimal hyperparameter combination, the adaptive weight, and a hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations to obtain multiple new hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the adaptive weight is determined by the maximum value of the distances between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the initial optimal hyperparameter combination and a boundary of the search range; Using the new multiple hyperparameter combinations as the multiple initial hyperparameter combinations, using the new initial optimal hyperparameter combination as the initial optimal hyperparameter combination, and iteratively performing operations for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination until a preset iteration condition is met; Determining the model parameters of the classification model according to the new initial optimal hyperparameter combination obtained when the preset iteration conditions are met, thereby obtaining an optimized classification model; The radar signal to be classified is classified using the optimized classification model to obtain a classification result of the radar signal.

2. The radar signal classification method according to claim 1, characterized in that: The updating of the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination according to the initial optimal hyperparameter combination, the adaptive weight, and the updated hyperparameter combination selected from the multiple initial hyperparameter combinations to obtain multiple new hyperparameter combinations and a new initial optimal hyperparameter combination includes: updating the multiple initial hyperparameter combinations according to the initial optimal hyperparameter combination, the adaptive weight, and the target hyperparameter combination and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations to obtain multiple hyperparameter combinations updated for the first time; Calculating the fitness values ​​of the classification models corresponding to the first updated plurality of hyperparameter combinations using the radar signal sample data set, and forming a fitness value set with the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination, taking the hyperparameter combination corresponding to the largest fitness value in the fitness value set as the initial optimal hyperparameter combination, to obtain the initial optimal hyperparameter combination after the first update; updating the multiple hyperparameter combinations updated for the first time according to the initial optimal hyperparameter combination after the first update, a rotation matrix having the same dimension as the initial optimal hyperparameter combination after the first update, and a hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the first time, to obtain multiple hyperparameter combinations updated for the second time; Calculating the fitness values ​​of the classification models corresponding to the plurality of hyperparameter combinations updated for the second time using the radar signal sample data set, and forming a fitness value set together with the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination after the first update, and using the hyperparameter combination corresponding to the largest fitness value in the fitness value set as the initial optimal hyperparameter combination, to obtain the initial optimal hyperparameter combination after the second update; updating the multiple hyperparameter combinations updated for the second time according to the initial optimal hyperparameter combination after the second update, the current number of iterations, the preset number of iterations, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the second time to obtain multiple new hyperparameter combinations; The radar signal sample data set is used to calculate the fitness values ​​of the classification models corresponding to each of the new multiple hyperparameter combinations, and the fitness values ​​of the classification models corresponding to the initial optimal hyperparameter combination after the second update are combined into a fitness value set. The hyperparameter combination corresponding to the largest fitness value in the fitness value set is used as the initial optimal hyperparameter combination to obtain a new initial optimal hyperparameter combination.

3. The radar signal classification method according to claim 2, characterized in that: The updating of the multiple initial hyperparameter combinations according to the initial optimal hyperparameter combination, the adaptive weight, and the target hyperparameter combination and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations to obtain multiple hyperparameter combinations updated for the first time includes: Selecting a target hyperparameter combination and a hyperparameter combination to be updated from the multiple initial hyperparameter combinations; Determining the adaptive weight according to the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the maximum value of the distance between the initial optimal hyperparameter combination and the boundary of the search range; updating the hyperparameter combination to be updated according to the initial optimal hyperparameter combination, the adaptive weight, the target hyperparameter combination, and the hyperparameter combination to be updated, and determining an updated first hyperparameter combination; Applying the updated first hyperparameter combination to replace the hyperparameter combination to be updated in the multiple initial hyperparameter combinations, and determining multiple hyperparameter combinations after replacement; Taking the replaced multiple hyperparameter combinations as the multiple initial hyperparameter combinations, selecting a target hyperparameter combination from the multiple initial hyperparameter combinations, and selecting a target updated hyperparameter combination from other initial hyperparameter combinations of the multiple initial hyperparameter combinations except the updated first hyperparameter combination; Taking the target update hyperparameter combination as the hyperparameter combination to be updated, iteratively perform operations for determining the adaptive weight, determining the first hyperparameter combination after update, determining the multiple hyperparameter combinations after replacement, and selecting the target hyperparameter combination and the target update hyperparameter combination until a first preset iteration condition is met, and the multiple hyperparameter combinations after replacement obtained in the current iteration round when the first preset iteration condition is met are used as the multiple hyperparameter combinations for the first update.

4. The radar signal classification method according to claim 2, characterized in that: The method of updating the multiple hyperparameter combinations updated for the first time according to the initial optimal hyperparameter combination after the first update, a rotation matrix having the same dimension as the initial optimal hyperparameter combination, and a hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the first time to obtain multiple hyperparameter combinations updated for the second time includes: Selecting a hyperparameter combination to be updated from the multiple hyperparameter combinations updated for the first time to obtain a first target updated hyperparameter combination; updating the first target update hyperparameter combination according to the rotation matrix, the first target update hyperparameter combination, and the initial optimal hyperparameter combination after the first update to determine an updated first target hyperparameter combination; Applying the updated first target hyperparameter combination to replace the first target updated hyperparameter combination in the plurality of hyperparameter combinations updated for the second time, to determine a plurality of first hyperparameter combinations; Taking the multiple first hyperparameter combinations as the multiple initial hyperparameter combinations, selecting a hyperparameter combination to be updated from other initial hyperparameter combinations of the multiple initial hyperparameter combinations except the updated first target hyperparameter combination, to obtain a new first target updated hyperparameter combination; Taking the new first target update hyperparameter combination as the first target update hyperparameter combination, iteratively execute the operation of determining the updated first target hyperparameter combination, the operation of determining multiple first hyperparameter combinations, and the operation of selecting a new first target update hyperparameter combination until the second preset iteration condition is met, and the multiple first hyperparameter combinations obtained in the current iteration round when the second preset iteration condition is met are used as the multiple hyperparameter combinations for the second update.

5. The radar signal classification method according to claim 2, characterized in that: The method of updating the multiple hyperparameter combinations updated for the second time according to the initial optimal hyperparameter combination after the second update, the current number of iterations, the preset number of iterations, and the hyperparameter combination to be updated selected from the multiple hyperparameter combinations updated for the second time to obtain multiple new hyperparameter combinations includes: Obtaining a second target updated hyperparameter combination by selecting a hyperparameter combination to be updated from the plurality of hyperparameter combinations updated for the second time; Determine a third target update hyperparameter combination according to the second target update hyperparameter combination, the initial optimal hyperparameter combination after the second update, the current number of iterations, and the preset number of iterations; Determine a fourth target updated hyperparameter combination according to the third target updated hyperparameter combination, the preset number of iterations, the current number of iterations, and the initial optimal hyperparameter combination after the second update; Applying the fourth target to update the hyperparameter combination, replacing the second target to update the hyperparameter combination in the plurality of hyperparameter combinations updated for the second time, and determining a plurality of second hyperparameter combinations; Taking the multiple second hyperparameter combinations as the multiple initial hyperparameter combinations, selecting a hyperparameter combination to be updated from other initial hyperparameter combinations of the multiple initial hyperparameter combinations except the fourth target updated hyperparameter combination, to obtain a new fourth target updated hyperparameter combination; Taking the new fourth target updating hyperparameter combination as the second target updating hyperparameter combination, iteratively perform the operations of determining the third target updating hyperparameter combination, determining the fourth target updating hyperparameter combination, determining multiple second hyperparameter combinations, and selecting a new fourth target updating hyperparameter combination until the third preset iteration condition is met, and the multiple second hyperparameter combinations obtained in the current iteration round when the third preset iteration condition is met are used as the multiple hyperparameter combinations for the third update.

6. The radar signal classification method according to claim 1, characterized in that: The optimized classification model includes an input layer, a convolutional neural network, a long short-term memory network, a self-attention layer, a linear layer, a fourth activation function layer and an output layer connected in sequence; The convolutional neural network includes a first convolution layer, a normalization layer, a first activation function layer, a second convolution layer, a dropout layer, a second activation function layer, a third convolution layer, a maximum pooling layer and a third activation function layer connected in sequence.

7. A radar signal classification device, characterized in that: The classification device comprises: An initial parameter determination module is configured to determine a plurality of initial hyperparameter combinations based on respective search ranges of the plurality of hyperparameters; each of the initial hyperparameter combinations includes the same number of hyperparameters; different initial hyperparameter combinations have different values ​​for at least some of the hyperparameters; each initial hyperparameter combination corresponds to a classification model; the classification model includes a convolutional neural network and a long short-term memory network; An optimal parameter determination module is used to calculate the fitness value of the classification model corresponding to each initial hyperparameter combination using the radar signal sample data set, and to use the initial hyperparameter combination corresponding to the maximum fitness value as the initial optimal hyperparameter combination; a parameter combination updating module, configured to update the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination according to the initial optimal hyperparameter combination, the adaptive weight, and the hyperparameter combination to be updated selected from the multiple initial hyperparameter combinations, to obtain multiple new hyperparameter combinations and a new initial optimal hyperparameter combination; wherein the adaptive weight is determined by the maximum value of the distance between the initial optimal hyperparameter combination, the hyperparameter combination to be updated, and the initial optimal hyperparameter combination and the boundary of the search range; an optimal parameter iteration module, configured to use the new multiple hyperparameter combinations as the multiple initial hyperparameter combinations, use the new initial optimal hyperparameter combination as the initial optimal hyperparameter combination, and iteratively perform operations for updating the multiple initial hyperparameter combinations and the initial optimal hyperparameter combination until a preset iteration condition is met; A classification model determination module is used to determine the model parameters of the classification model according to the new initial optimal hyperparameter combination obtained when the preset iteration conditions are met, so as to obtain an optimized classification model; The radar signal classification module is used to classify the radar signal to be classified using the optimized classification model to obtain the classification result of the radar signal.

8. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that Executable instructions are stored thereon, which, when executed by a processor, enable the processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images