Radar operating mode recognition method and system based on human-computer interaction, and electronic device

Abstract

The application provides a radar working mode recognition method and system based on human-computer interaction and an electronic device, the method comprising: training a classification algorithm model based on data in a database to obtain an initial radar working mode recognition model; receiving radar working parameters to be recognized input by a user, outputting a recognition result to a user interface through the initial radar working mode recognition model; receiving feedback information provided by the user through the user interface, converting the feedback information into format data processable by the system, and using the feedback information to update the database; based on the database, calculating the loss of the initial radar working mode recognition model, and adjusting the parameters of the initial radar working mode recognition model according to the loss. The application can not only efficiently and accurately recognize the radar working mode, but also can interact with the user, and has the ability of incremental learning, and can adapt to the changing radar signal environment and user demand.

Description

Technical Field

[0001] This application belongs to the field of radar signal recognition and processing technology, and relates to a radar operating mode recognition method, system and electronic equipment based on human-computer interaction. Background Technology

[0002] Radar operating mode recognition is a key technology in the field of radar signal processing. Its main applications include electronic reconnaissance and intelligence analysis. In electronic reconnaissance, this technology is used to analyze the characteristics of enemy radar signals and determine the operating mode and tactical intent of the radar. With the continuous development of radar technology and the emergence of new types of radars, traditional radar operating mode recognition methods mainly rely on radar signal preprocessing, feature extraction, and classification algorithms. These methods typically only achieve operating mode recognition and lack human-computer interaction with the user. Traditional radar operating mode recognition methods cannot automatically convert user data into data required by the technical system, and the processing of user data often requires specialized technical expertise. Furthermore, traditional radar operating mode recognition methods usually use offline models for training and recognition. These offline models are trained before deployment and cannot be automatically and dynamically adjusted and optimized based on actual user data, thus failing to accurately identify the operating modes of new radars and limiting their application in complex electromagnetic environments. Summary of the Invention

[0003] The purpose of this application is to provide a radar operating mode recognition method, system, and electronic device based on human-computer interaction, which solves the problems that existing radar operating mode recognition methods cannot realize human-computer interaction with users, cannot accurately identify the operating modes of new radars, and cannot automatically and dynamically adjust and optimize based on actual user data.

[0004] In a first aspect, this application provides a radar operating mode recognition method based on human-computer interaction, the method comprising:

[0005] The classification algorithm model is trained based on the data in the database to obtain an initial radar operating mode recognition model; wherein, the database includes formatted radar operating mode parameter information;

[0006] Receive the radar operating parameters to be identified input by the user, and output the identification result to the user interface through the initial radar operating mode identification model;

[0007] The system receives feedback information provided by the user through the user interface, converts the feedback information into data in a format that the system can process, and uses it to update the database; wherein, the feedback information includes correction information for the recognition result or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language form;

[0008] Based on the database, the loss of the initial radar operating mode recognition model is calculated, and the parameters of the initial radar operating mode recognition model are adjusted according to the loss to obtain the updated radar operating mode recognition model.

[0009] In one implementation, before calculating the loss of the initial radar operating mode recognition model based on the database, the process includes:

[0010] Based on the database, the radar operating mode parameter information in the database is preprocessed to make the database match the input format of the initial radar operating mode recognition model; wherein, the preprocessing includes denoising, pulse extraction and parameter measurement of the data in the database.

[0011] In one implementation, the step of calculating the loss of the initial radar operating mode recognition model based on the database, and adjusting the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model includes:

[0012] The radar operating mode parameter information in the database is divided into historical category data and newly added category data. The initial radar operating mode recognition model is used as the teacher model to train the historical category data and generate labels.

[0013] Construct a student model and calculate the combined loss of the student model;

[0014] The weights of the student model are updated based on the combined loss to obtain the updated radar operating mode recognition model.

[0015] In one implementation, calculating the combined loss of the student model includes:

[0016] Based on the newly added category data, a weighted cross-entropy loss is calculated on the newly added category data to obtain a first loss value; wherein, the sample weight of the newly added category data is set inversely proportional to the frequency of the newly added category.

[0017] The second loss value is obtained by calculating the KL divergence distillation loss based on the historical category data and the output probability distribution of the teacher model and student model.

[0018] The combined loss of the student model is obtained by weighted summation of the first loss value and the second loss value.

[0019] In one implementation, updating the weights of the student model based on the combined loss includes:

[0020] The number of output layer neurons in the student model is dynamically expanded according to the category of the newly added radar working mode. During training, the weights of the newly added output layer neurons participate in the complete backpropagation, and the weights of the original output layer neurons also participate in the update. However, the gradients of the original output layer neurons are constrained by the knowledge distillation loss to retain historical knowledge. The weights of the output layer neurons are initialized using a Gaussian distribution.

[0021] In one implementation, adjusting the parameters of the initial radar operating mode recognition model based on the loss includes:

[0022] Based on the aforementioned loss, when the amount of feedback information exceeds a preset threshold, an incremental learning optimization algorithm is used to dynamically adjust the parameters of the initial radar working mode recognition model, so as to retain old knowledge while learning new knowledge.

[0023] In one implementation, converting the feedback information into system-processable format data and using it to update the database includes:

[0024] The prompt word is input into the large language model, which generates a corresponding database operation statement based on the prompt word and the feedback information, and executes the database operation statement to update the database.

[0025] The prompts include database table structure and field name information.

[0026] In one implementation, the classification algorithm model may include fully connected neural networks, CNNs, RNNs, LSTMs, or ResNets, etc.

[0027] Secondly, a radar operating mode recognition system based on human-computer interaction is provided, the system comprising:

[0028] The algorithm module is configured to train the classification algorithm model based on data in the database to obtain an initial radar operating mode recognition model; wherein, the database includes formatted radar operating mode parameter information;

[0029] The human-machine intelligent interaction module is configured to receive radar operating parameters to be identified input by the user, output the identification result to the user interface through the initial radar operating mode identification model, and receive feedback information provided by the user through the user interface; wherein, the feedback information includes correction information for the identification result or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language form;

[0030] The intelligent incremental learning module is configured to convert the feedback information into data in a format that the system can process and use it to update the database, calculate the loss of the initial radar operating mode recognition model based on the database, and adjust the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model.

[0031] Thirdly, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the above-described method.

[0032] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the steps of the above-described method.

[0033] Fifthly, a computer program product is provided, comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of an implementation method.

[0034] As described above, the radar operating mode recognition method, system, and electronic device based on human-computer interaction described in this application have the following beneficial effects:

[0035] This application receives radar operating parameters input by the user and outputs the recognition result to the user interface through the initial radar operating mode recognition model. It also receives feedback information provided by the user through the user interface, including correction information for the recognition result or newly entered radar operating mode parameter information. The newly entered radar operating mode parameter information is input in natural language and converted into a system-processable format for updating the database. This not only efficiently and accurately identifies radar operating modes but also enables human-computer interaction, automatically converting the user's natural language data into database operation statements required by the system. Furthermore, this application dynamically adjusts the parameters of the initial radar operating mode recognition model by adjusting the loss of the initial radar operating mode recognition model to dynamically update the model, increasing incremental learning capabilities to adapt to constantly changing radar signal environments and user needs. This solves the problems of existing radar operating mode recognition methods, such as the inability to achieve human-computer interaction, the inability to accurately identify the operating modes of new radars, and the inability to dynamically adjust and optimize based on actual user data. Attached Figure Description

[0036] Figure 1 The diagram shown is a flowchart of a radar operating mode recognition method based on human-computer interaction provided in an embodiment of this application.

[0037] Figure 2The flowchart shown is a method provided in this application embodiment for calculating the loss of an initial radar operating mode recognition model based on a database, and adjusting the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model.

[0038] Figure 3 The flowchart shown is a method for calculating the combined loss of a student model provided in an embodiment of this application.

[0039] Figure 4 The diagram shown is a structural diagram of a radar operating mode recognition system based on human-computer interaction provided in an embodiment of this application.

[0040] Figure 5 The diagram shown is a structural diagram of an electronic device provided in an embodiment of this application.

[0041] Explanation of reference numerals on the accompanying drawings:

[0042] 1. Algorithm module; 2. Human-computer intelligent interaction module; 3. Intelligent incremental learning module; 21. GPU; 22. CPU; 23. Memory; 24. Storage device; 25. Communication interface. Detailed Implementation

[0043] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0044] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The illustrations only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0045] In the field of electronic reconnaissance and intelligence analysis, radar operating mode recognition technology plays a crucial role in signal translation and intent prediction. Electronic reconnaissance systems detect radar signals from potential threats through platforms such as satellites, aircraft, and ground stations. Pattern recognition technology first performs in-depth processing on the captured raw signals, measuring parameters such as radio frequency, repetition rate, pulse width, and beam scanning mode. Intelligent algorithms then determine the specific operating mode of the radar, such as search or tracking. This recognition capability transforms signal characteristics into dynamic intelligence with tactical significance: for example, a radar previously in search mode suddenly switching to tracking mode indicates that its platform may have detected and locked onto a specific target, foreshadowing an impending escalation of hostile actions; by monitoring the patterns and changes in all radar modes in a given area over a long period, intelligence personnel can outline the enemy's defense system layout. Therefore, radar operating mode recognition is not only the end point of signal analysis but also the starting point of tactical intent analysis, providing crucial intelligence support for decision-making by "seeing through the phenomena to the essence."

[0046] To address the aforementioned technical problems, this application provides a radar operating mode recognition method, system, and electronic device based on human-computer interaction in the following embodiments. This application receives radar operating parameters input by a user, outputs the recognition result to the user interface through the initial radar operating mode recognition model, and receives feedback information provided by the user through the user interface. The feedback information includes correction information for the recognition result or newly entered radar operating mode parameter information. The newly entered radar operating mode parameter information is input in natural language form and converted into system-processable format data for updating the database. This not only efficiently and accurately identifies radar operating modes but also enables human-computer interaction, automatically converting the user's natural language data into database operation statements required by the system. Furthermore, this application dynamically adjusts the parameters of the initial radar operating mode recognition model by adjusting the loss of the initial radar operating mode recognition model to dynamically update the model, increasing incremental learning capabilities to adapt to constantly changing radar signal environments and user needs. This solves the problems of existing radar operating mode recognition methods failing to achieve human-computer interaction, accurately identify new radar operating modes, and dynamically adjust and optimize based on actual user data.

[0047] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0048] like Figure 1 As shown in the figure, this embodiment provides a radar operating mode recognition method based on human-computer interaction, the method including:

[0049] Step 100: Train the classification algorithm model based on the data in the database to obtain the initial radar working mode recognition model.

[0050] The database includes formatted radar operating mode parameter information.

[0051] Specifically, based on the generated database, instantiated data is abstracted, a dataset for training the radar operating mode recognition model is constructed, a radar operating mode recognition model structure that conforms to the database size is designed, and an initial radar operating mode recognition model is trained.

[0052] Furthermore, the method for constructing the training dataset includes: extracting radar operating mode parameters from the database, performing structured feature decomposition on each type of radar operating mode parameter, generating a data matrix from the structured feature decomposition radar operating mode parameters, and the data matrix constituting the training dataset.

[0053] For example, the carrier frequency type includes three types: fixed carrier frequency, pulse group agility, or inter-pulse agility. The frequency values ​​in the database are all ranges, falling between the minimum and maximum frequency values. If the carrier frequency type is fixed carrier frequency, both the minimum and maximum frequency values ​​are set to fixed values. If the carrier frequency type is pulse group agility or inter-pulse agility, values ​​are assigned according to the actual range.

[0054] For example, the input matrix dimension of this data matrix is The output matrix has a dimension of ,in, This represents the total number of radar operating mode samples. For the characteristic dimensions of radar operating mode parameters, This represents the total number of work mode categories.

[0055] Furthermore, the radar operating mode recognition model structure can be a fully connected network.

[0056] Furthermore, the input layer of this fully connected network receives The feature matrix and hidden layers are set as a multi-layer network structure, with each layer followed by a ReLU activation function. Batch normalization layers are added after each hidden layer to accelerate training and improve model performance. The output layer is set as a fully connected classification layer with K nodes. The Softmax activation function is used to convert the output into a probability distribution, resulting in K output categories.

[0057] In some embodiments, radar operating mode parameters include carrier frequency, pulse width, pulse repetition interval, and pulse amplitude.

[0058] In some embodiments, the value types of radar operating mode parameter information include, but are not limited to, strings, numerical values, or numerical ranges.

[0059] In some embodiments, a high-speed signal processing pipeline can be built using a field-programmable gate array (FPGA) to extract radar operating mode parameter information in real time.

[0060] In some embodiments, radar operating mode parameter information can be collected through a distributed sensor network.

[0061] This embodiment converts radar operating mode parameter information into a clear, machine-readable format and splits it into text blocks with metadata tags according to semantics, so that the system can retrieve and understand it efficiently. The clear, machine-readable format is such as Markdown or JSON.

[0062] Specifically, the classification algorithm model includes fully connected neural networks, CNNs, RNNs, LSTMs, or ResNets.

[0063] The classification algorithm model in this embodiment adopts a deep learning model, which can significantly outperform traditional algorithms in complex classification tasks through automatic feature learning, multi-layer nonlinear transformation and end-to-end optimization, especially in the field of unstructured data such as images, text and speech.

[0064] In some embodiments, the method for generating the database includes: collecting radar operating mode information parameters, generating a radar operating mode knowledge base, and formatting the content of the knowledge base before importing it into the database. Subsequent data operations can be performed using a database operation language. This database operation language is categorized into data query language, data manipulation language, data definition language, and data control language.

[0065] Specifically, Data Query Language (DML) is primarily used to retrieve data from a database, making data queries intuitive and efficient. Data Manipulation Language (DML) is used for operations such as adding, deleting, and updating data in the database. DML is often used in conjunction with DML, first retrieving data and then manipulating it as needed. Data Definition Language (DML) is used to define and manage the structure and objects of a database. Through DML, users can create new databases, tables, indexes, etc., and can also modify or delete existing database structures. Data Control Language (DCL) is primarily used to control database access permissions and security settings, ensuring data security and privacy, and preventing unauthorized access and modification.

[0066] Optionally, to ensure system security and data integrity, after the large language model generates database operation statements, the system can further introduce a post-processing verification module to perform multi-layered verification on the statements: First, through SQL syntax tree parsing, the syntactic structure of the generated statements is analyzed for legality, eliminating illegal instructions with format errors or containing dangerous keywords (such as DROP, DELETE, TRUNCATE); second, combined with the current user's permission configuration, fine-grained permission checks are performed to ensure that the user can only operate on authorized database tables and fields; finally, the statements to be executed are simulated in an isolated sandbox environment to verify their actual impact (such as whether insertion and update meet expectations), and only after confirming that there is no abnormal behavior are they allowed to be submitted to the real database for execution. This post-processing verification mechanism effectively prevents the risk of data leakage or damage caused by the illusion of the large language model, user mis-input, or malicious instructions, improving the robustness and security of the system in human-computer interaction scenarios.

[0067] In this embodiment, the database supports data import and export operations. The initial radar operating mode recognition model uses these data for training and iterative optimization.

[0068] Step 200: Receive the radar operating parameters to be identified input by the user, and output the identification results to the user interface through the initial radar operating mode identification model.

[0069] Specifically, after the user inputs information such as the radar operating parameters to be identified, the user interface receives the data and sends it to the system backend. The system backend calls the algorithm module and the human-machine intelligent interaction module to output the radar operating mode recognition result and display it on the user interface.

[0070] Specifically, in actual use, the system's software interface (user interface) will display the radar operating mode recognition results and the input radar operating parameters to be identified.

[0071] Step 300: Receive feedback information provided by the user through the user interface, convert the feedback information into data in a format that the system can process, and use it to update the database; wherein, the feedback information includes correction information for the identification result or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language form.

[0072] For example, the new radar operating mode parameter information input by the user, mainly in natural language form, is radar model xx, carrier frequency 1000MHz, carrier frequency type agile, pulse period 100us, and pulse repetition interval 1000us.

[0073] Specifically, before calculating the loss of the initial radar operating mode recognition model based on the database, the process includes: preprocessing the radar operating mode parameter information in the database to make the database match the input format of the initial radar operating mode recognition model; wherein, the preprocessing includes denoising, pulse extraction and parameter measurement of the data in the database.

[0074] This embodiment ensures the consistency of data in the database by preprocessing the database, improves the data input quality of the initial radar operating mode recognition model, and enables the initial radar operating mode recognition model to learn new radar operating mode parameter information more efficiently and stably.

[0075] Specifically, if a user finds an error in the operating mode recognition result displayed on the user interface, they can manually modify the recognition result and send the modified data to the system backend. Alternatively, if a user discovers a new type of radar or a new operating mode of an existing radar, the user can enter the new radar operating mode parameter information on the user interface. After the user completes the modification of the recognition result or adds the new radar operating mode parameter information, the system will preprocess the radar operating mode parameter information in the database to match the input format of the initial radar operating mode recognition model.

[0076] In some embodiments, the new radar operating mode and the existing radar operating modes both include search mode, tracking mode, imaging mode, weather detection mode, navigation mode, etc.

[0077] For example, in search mode, the radar can scan a wide area to detect and locate potential targets. In tracking mode, once the radar detects a target, it can switch to tracking mode to continuously monitor the target's position, velocity, and acceleration. In imaging mode, synthetic aperture radar operates in this mode to acquire high-resolution ground images. In weather detection mode, weather radar detects and displays aerial weather targets, such as clouds and precipitation, helping pilots avoid areas of severe weather and ensuring flight safety. In navigation mode, the radar can provide navigation information for aircraft or ships.

[0078] This embodiment supports the use of natural language in the user interface, eliminating the need for users to master professional database commands. This embodiment also realizes human-computer interaction.

[0079] In some embodiments, the feedback information can be converted into database operation statements using a large language model, and the database operation statements can be executed to update the database.

[0080] Specifically, converting the feedback information into format data that can be processed by the system and used to update the database includes: inputting the prompt words into the large language model, which generates corresponding database operation statements based on the prompt words and the feedback information, and executing the database operation statements to update the database.

[0081] Among them, the prompt words include database table structures and field name information.

[0082] Exemplarily, assume the prompt words are "Please generate a database operation statement applicable to the MySQL database. The table name of the MySQL database is student, which contains the primary key id of type INT and name of type VARCHAR(100). Now want to add a piece of data with id of 10 and name of Zhang San". The large language model can generate the statement "insert into student VALUES(10,'Zhang San')". After the system receives this statement, it sends it to the database for execution, and the database will add a new piece of data to the database.

[0083] In some embodiments, the large language model can be DeepSeek or Qianwen, etc.

[0084] In this embodiment, by adding prompt words to the large language model, the accuracy of generating database operation statements can be improved, making the generated database operation statements more in line with requirements.

[0085] Step 400: Specifically, as Figure 2 shown, based on the database, calculate the loss of the initial radar working mode recognition model, and adjust the parameters of the initial radar working mode recognition model according to the loss to obtain an updated radar working mode recognition model, including:

[0086] Step 401: Divide the radar working mode parameter information in the database into historical category data and newly added category data, and use the initial radar working mode recognition model as the teacher model to train the historical category data and generate labels.

[0087] Step 402: Construct a student model and calculate the combined loss of the student model.

[0088] Among them, as Figure 3 shown, calculating the combined loss of the student model includes:

[0089] 4021: Based on the newly added category data, calculate the weighted cross-entropy loss for the newly added category data to obtain a first loss value; among them, the sample weights of the newly added category data are set inversely proportional to the newly added category frequency.

[0090] The sample weights for newly added category data are set inversely proportional to the frequency of the newly added category. This means that the weight is calculated based on the number of samples for each newly added category. The fewer samples a newly added category has, the higher its weight is assigned. Specifically, the weight is inversely proportional to the number of samples for the newly added category.

[0091] Specifically, the formula for expressing the first loss value is as follows: ,in, A value greater than 0 indicates the weight of the newly added category data belonging to the k-th class. These are real tags, with values ​​of 0 or 1. This is the predicted probability distribution output by the student model for the newly added radar operating mode. This is the first loss value.

[0092] 4022: Based on the historical category data and the KL divergence distillation loss calculated from the output probability distribution of the teacher model and student model, the second loss value is obtained.

[0093] Specifically, the formula for expressing this second loss value is as follows:

[0094] ,in, This represents the output probability distribution of the teacher model at temperature T. This represents the output probability distribution of the student model at temperature T. This represents the second loss value. This is a newly added radar operating mode for input.

[0095] Used to measure two probability distributions and The difference between them is expressed by the following formula:

[0096] ,in, For teacher models in categories The probability value on, For student models in categories The probability value above.

[0097] In weighted cross-entropy loss, the weight of each class is typically inversely proportional to the number of samples in that class; that is, the fewer samples a class has, the greater its weight in the loss function. Due to the imbalance of classes in the dataset, weighted cross-entropy loss is used to reduce the sensitivity of the loss function to different classes by assigning different weights to each class, thereby reducing the risk of overfitting in the radar working pattern recognition model.

[0098] 4023: The first loss value and the second loss value are weighted and summed to obtain the combined loss of the student model.

[0099] The specific formula for calculating the combined loss is: L_total=α* +β* ,in, This represents the first loss value. L represents the second loss value, α and β represent the weights, α+β=1, and L_total represents the combined loss.

[0100] In some embodiments, the value of α is generally set to 0.5 or higher to ensure that old knowledge is not forgotten while learning new knowledge.

[0101] In some embodiments, the values ​​of α and β are determined through grid search optimization.

[0102] Step 403: Update the weights of the student model based on the combined loss to obtain the updated radar working mode recognition model.

[0103] The teacher model is the radar operating mode recognition model before the update, and the student model is the radar operating mode recognition model after the update.

[0104] Specifically, updating the weights of the student model based on the combined loss includes: the number of output layer neurons of the student model is dynamically expanded according to the category of the newly added radar working mode; during training, the weights of the newly added output layer neurons participate in the complete backpropagation, and the weights of the original output layer neurons also participate in the update, but the gradients of the original output layer neurons are constrained by the knowledge distillation loss to retain historical knowledge; wherein, the weights of the output layer neurons are initialized using a Gaussian distribution.

[0105] Full backpropagation refers to the entire process in neural network training where, starting from the output layer, the gradient of the loss function with respect to all network parameters (including the weights of newly added and existing output layer neurons) is calculated layer by layer using the chain rule, and these parameters are updated using gradient descent. This process ensures that all weights are optimized based on the current loss function, while using knowledge distillation loss to constrain the gradients of existing output layer neurons, thus balancing new task learning with the preservation of historical knowledge.

[0106] In this embodiment, the weights of the original output layer neurons are constrained by distillation loss to retain old knowledge, while the weights of the newly added output layer neurons are dominated by cross-entropy loss to learn new category features.

[0107] Furthermore, when unidentified new radar operating mode parameter information is detected in the database, k output neurons are added to the output layer of the student model according to the number k of newly added radar operating mode categories. The connection weight matrix W_new between the newly added output layer neurons and the hidden layer is initialized using a zero-mean Gaussian distribution. The probability density function of the zero-mean Gaussian distribution satisfies: W_new ~ N(0, σ²), where the standard deviation σ ranges from 0.01 ≤ σ ≤ 0.05 to ensure that the newly added weights have minimal perturbation and avoid destroying the learned feature representation. The weights of the newly added output layer neurons participate in backpropagation optimization, learning the features of the unidentified new radar operating modes through weighted cross-entropy loss. In the incremental learning iteration, the weights of the original output layer neurons and the newly added output layer neurons are simultaneously optimized through a combined loss function until the training accuracy of the newly added radar operating mode categories reaches or approaches the original level, at which point the iteration stops.

[0108] In some embodiments, σ ​​is preferably 0.02.

[0109] For example, when two new radar operating modes are detected, such as "low probability of intercept mode" and "jamming countermeasure mode", the number of neurons in the original output layer is expanded from k=8 to k'=10. The weights of the new neurons are initialized with σ=0.02, and the bias term is initialized to 0.

[0110] This embodiment can update itself by gradually receiving new data or new tasks without retraining the entire dataset.

[0111] Specifically, the parameters of the initial radar operating mode recognition model are adjusted based on the loss, including: based on the loss, when the amount of feedback information exceeds a preset threshold, the parameters of the initial radar operating mode recognition model are dynamically adjusted using an incremental learning optimization algorithm to retain old knowledge while learning new knowledge.

[0112] In some embodiments, the preset threshold can be dynamically adjusted according to the actual needs of those skilled in the art.

[0113] For example, the preset threshold can be set to 100.

[0114] In this embodiment, a preset threshold is set. Only when the amount of feedback information exceeds the preset threshold will the incremental learning optimization algorithm be activated to dynamically adjust the parameters of the initial radar working mode recognition model, reduce the number of training sessions, and retain old knowledge while learning new knowledge.

[0115] Specifically, for incremental learning optimization algorithms, the Levenberg-Marquardt (LM) optimization algorithm can be chosen. It balances global search capability and local convergence speed by dynamically adjusting the damping factor, exhibiting excellent performance on small datasets and medium-sized networks. If the accumulated data is large enough, optimization algorithms such as Adam can replace the Levenberg-Marquardt optimization algorithm.

[0116] In this embodiment, the classification loss focuses on learning new operating mode categories defined by natural language interaction, while the distillation loss ensures that while expanding the types of radar operating modes that can be recognized, it does not forget the knowledge of old modes that have already been acquired. The updated radar operating mode recognition model is seamlessly deployed back to the radar operating mode recognition module, making it recognize a wider range of radar operating modes while maintaining the recognition accuracy of existing operating modes.

[0117] Through the above technical solution, the radar working mode recognition accuracy of this embodiment can reach more than 80%. For example, when the total number of radar working mode samples is 100,000, the radar working mode recognition accuracy can reach about 90%.

[0118] The scope of protection of the radar working mode recognition method based on human-computer interaction described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0119] This application also provides a radar operating mode recognition system based on human-computer interaction. The radar operating mode recognition system based on human-computer interaction can implement the radar operating mode recognition method based on human-computer interaction of this application. However, the implementation device of the radar operating mode recognition method based on human-computer interaction of this application includes, but is not limited to, the structure of the radar operating mode recognition system based on human-computer interaction listed in this embodiment. All structural modifications and substitutions of the prior art made according to the principles of this application are included within the protection scope of this application.

[0120] like Figure 4As shown, this embodiment provides a radar operating mode recognition system based on human-computer interaction. The system includes: an algorithm module 1, configured to train a classification algorithm model based on data in a database to obtain an initial radar operating mode recognition model; wherein the database includes formatted radar operating mode parameter information; a human-computer intelligent interaction module 2, configured to receive radar operating parameters to be recognized input by the user, output recognition results to the user interface through the initial radar operating mode recognition model, and receive feedback information provided by the user through the user interface; wherein the feedback information includes correction information for the recognition results or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language; and an intelligent incremental learning module 3, configured to convert the feedback information into system-processable format data for updating the database, and based on the database, calculate the loss of the initial radar operating mode recognition model, and adjust the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model.

[0121] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses, modules, or units may be electrical, mechanical, or other forms.

[0122] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0123] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] In one exemplary embodiment, an electronic device is provided, such as Figure 5 As shown, the electronic device includes a GPU 21, a CPU 22, a memory 23, a storage device 24, and a communication interface 25.

[0125] GPU stands for Graphics Processing Unit; CPU stands for Central Processing Unit.

[0126] In this invention, the GPU is used for radar working mode algorithm training and interface display, the CPU is used for calculation and instruction control, the memory is used for temporary data storage, the communication interface and communication bus are used for communication and data interaction, and the memory is used for data storage related to the working mode algorithm.

[0127] The graphics processing unit (GPU) and central processing unit (CPU) possess data processing capabilities, enabling them to read data or programs from memory, train and optimize algorithms, and implement the algorithms and software of this invention. The memory can store program code and data; the program code may include instructions for implementing the method of this invention, and the data may include data used and generated during the execution of the method of this invention. The memory may include random access memory (RAM), read-only memory (ROM), etc., and may include hard disk drives (HDDs), solid-state drives (SSDs), etc. The communication interface and communication bus can be used for communication between applications or between electronic devices. In specific implementations, the communication interface may include wired and wireless communication interfaces, and the communication bus may include a system bus, I / O bus, PCIe bus, etc.

[0128] In one exemplary embodiment, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0129] Electronic devices can be, but are not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, smart cars, etc., while portable wearable devices can be smartwatches and smart bracelets, etc. It is understood that this method can also be applied to systems that include electronic devices and servers, and is implemented through the interaction between electronic devices and servers.

[0130] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above-described method embodiments.

[0131] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0132] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0133] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random-access memory (ReRAM), magnetic random-access memory (MRAM), ferroelectric random-access memory (FRAM), phase-change memory (PCM), graphene memory, etc. Volatile memory can include random-access memory (RAM) or cache, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic units, artificial intelligence (AI) processors, etc., and are not limited to these.

[0134] The descriptions of the processes or structures corresponding to the various figures above each have their own emphasis. For parts of a process or structure not described in detail, please refer to the relevant descriptions of other processes or structures. The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of this application. Therefore, any equivalent modifications or substitutions made by a person skilled in the art to the above embodiments without departing from the technical solutions and spirit disclosed in this application should fall within the protection scope of the claims of this application.

Claims

1. A radar operating mode recognition method based on human-computer interaction, characterized in that, The method includes: The classification algorithm model is trained based on the data in the database to obtain an initial radar operating mode recognition model; wherein, the database includes formatted radar operating mode parameter information; Receive the radar operating parameters to be identified input by the user, and output the identification result to the user interface through the initial radar operating mode identification model; The system receives feedback information provided by the user through the user interface, converts the feedback information into data in a format that the system can process, and uses it to update the database; wherein, the feedback information includes correction information for the recognition result or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language form; based on the database, the system calculates the loss of the initial radar operating mode recognition model, and adjusts the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model; The step of converting the feedback information into system-processable format data and using it to update the database includes: inputting a prompt word into a large language model, the large language model generating a corresponding database operation statement based on the prompt word and the feedback information, and executing the database operation statement to update the database; the prompt word includes database table structure and field name information; The step of calculating the loss of the initial radar operating mode recognition model based on the database, and adjusting the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model includes: dividing the radar operating mode parameter information in the database into historical category data and newly added category data; using the initial radar operating mode recognition model as a teacher model, training the historical category data and generating labels; constructing a student model and calculating the combined loss of the student model; and updating the weights of the student model according to the combined loss to obtain the updated radar operating mode recognition model. The calculation of the combined loss of the student model includes: calculating a weighted cross-entropy loss on the newly added category data based on the newly added category data to obtain a first loss value; wherein the sample weights of the newly added category data are set inversely proportional to the frequency of the newly added category; calculating a second loss value based on the historical category data and the KL divergence distillation loss based on the output probability distributions of the teacher model and the student model; and performing a weighted summation of the first loss value and the second loss value to obtain the combined loss of the student model. The step of updating the weights of the student model according to the combined loss includes: the number of output layer neurons of the student model is dynamically expanded according to the category of the newly added radar working mode; during the training process, the weights of the newly added output layer neurons participate in the complete backpropagation, and the weights of the original output layer neurons also participate in the update, but the gradients of the original output layer neurons are constrained by the knowledge distillation loss to retain historical knowledge; wherein, the weights of the number of output layer neurons are initialized using a Gaussian distribution.

2. The method according to claim 1, characterized in that, Before calculating the loss of the initial radar operating mode recognition model based on the database, the process includes: Based on the database, the radar operating mode parameter information in the database is preprocessed to make the database match the input format of the initial radar operating mode recognition model; wherein, the preprocessing includes denoising, pulse extraction and parameter measurement of the data in the database.

3. The method according to claim 1, characterized in that, The step of adjusting the parameters of the initial radar operating mode recognition model based on the loss includes: Based on the aforementioned loss, when the amount of feedback information exceeds a preset threshold, an incremental learning optimization algorithm is used to dynamically adjust the parameters of the initial radar working mode recognition model, so as to retain old knowledge while learning new knowledge.

4. The method according to claim 1, characterized in that, The classification algorithm model includes fully connected neural networks, CNNs, RNNs, LSTMs, or ResNets.

5. A radar operating mode recognition system based on human-computer interaction, characterized in that, The system includes: The algorithm module is configured to train the classification algorithm model based on data in the database to obtain an initial radar operating mode recognition model; wherein, the database includes formatted radar operating mode parameter information; A human-machine intelligent interaction module is configured to receive radar operating parameters to be identified input by the user, output identification results to the user interface through the initial radar operating mode recognition model, and receive feedback information provided by the user through the user interface; wherein, the feedback information includes correction information for the identification results or newly entered radar operating mode parameter information; the newly entered radar operating mode parameter information is input in natural language form; an intelligent incremental learning module is configured to convert the feedback information into data in a format that the system can process and use it to update the database, and based on the database, calculate the loss of the initial radar operating mode recognition model, and adjust the parameters of the initial radar operating mode recognition model according to the loss to obtain an updated radar operating mode recognition model; The step of converting the feedback information into system-processable format data and using it to update the database includes: inputting a prompt word into a large language model, the large language model generating a corresponding database operation statement based on the prompt word and the feedback information, and executing the database operation statement to update the database; the prompt word includes database table structure and field name information; The radar operating mode parameter information in the database is divided into historical category data and newly added category data. The initial radar operating mode recognition model is used as the teacher model to train the historical category data and generate labels. A student model is constructed, and the combined loss of the student model is calculated. The weights of the student model are updated according to the combined loss to obtain the updated radar operating mode recognition model. The calculation of the combined loss of the student model includes: calculating a weighted cross-entropy loss on the newly added category data based on the newly added category data to obtain a first loss value; wherein the sample weights of the newly added category data are set inversely proportional to the frequency of the newly added category; calculating a second loss value based on the historical category data and the KL divergence distillation loss based on the output probability distributions of the teacher model and the student model; and performing a weighted summation of the first loss value and the second loss value to obtain the combined loss of the student model. The step of updating the weights of the student model according to the combined loss includes: the number of output layer neurons of the student model is dynamically expanded according to the category of the newly added radar working mode; during the training process, the weights of the newly added output layer neurons participate in the complete backpropagation, and the weights of the original output layer neurons also participate in the update, but the gradients of the original output layer neurons are constrained by the knowledge distillation loss to retain historical knowledge; wherein, the weights of the number of output layer neurons are initialized using a Gaussian distribution.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

Images