A radar operating mode recognition method and device based on a predicted state representation

Abstract

The application discloses a radar working mode recognition method and device based on a predicted state representation, comprising the following steps: training a probability model in advance by using a statistical method; sequentially searching for possible working mode transition probabilities of each time step from the probability model according to known states and known modes, and calculating transition path lengths; determining the shortest path of mode transition and the corresponding path value in the entire time sliding window according to the possible mode transition probabilities of each time step and the corresponding mode transition path lengths; recording the above results, moving the sliding window down by one time step, and repeating the above process; and voting to determine the optimal mode recognition result of each time step according to the voting result. The application estimates the mode value of each time step by searching for the maximum a posteriori probability of the mode sequence under the condition of the known state sequence, and adjusts and optimizes the recognition effect by comparing the recognition results in different sliding windows.

Description

Technical Field

[0001] The present application relates to the field of radar technology, and in particular to a radar operating mode recognition method and device based on predicted state representation. Background Art

[0002] To address complex radar countermeasures in dynamic environments, deep understanding of emitters through radar behavior analysis is necessary. A key research topic is radar operating mode recognition. This refers to the process by which a computer, based on received radar signals and their parameters, analyzes and extracts the underlying characteristics of a target radar, identifying its operating mode. Radar operating mode recognition is the foundation of cognitive countermeasures and serves as the starting point for key cognitive countermeasure processes, including emitter threat estimation, behavioral intention inference, adaptive countermeasure strategy generation, and online jamming effect analysis. Compared to traditional emitter parameter analysis and extraction, radar operating modes must consider the interrelationships of more parameters. Different operating modes correspond to numerous complex parameter combinations, making them difficult to identify using parameter matching. The widespread application of machine learning methods in recent years has provided new approaches to pattern recognition.

[0003] Visnevski N., Syntactic Modeling of Multi-Function Radars [D], McMaster University, 2005; proposed a grammatical structure model for describing radar waveform sequences, achieving a formal representation of the rules governing radar signal generation and variation. This model uses "radar words" composed of several pulses (rather than single pulses as used in traditional analysis methods) as the minimum unit for analyzing and predicting radar sequence signal characteristics, effectively controlling the model's complexity. To implement the functions of each radar mode, multiple "radar words" are sequentially connected to form "radar phrases." A single radar operating mode may correspond to multiple "radar phrases," and a single "radar phrase" may also correspond to multiple operating modes. From the receiver's perspective, the radar continuously emits various "radar phrases" during its mission. These "radar phrases" are sequentially combined to form "radar sentences," providing a complete description of the radar's workflow. For example, each operating mode of the Mercury multi-function radar corresponds to several radar phrases, each consisting of four radar words. Because the grammatical model focuses on the combination of radar words and radar phrases, it also has the ability to represent radars with unknown parameters.

[0004] Littman M., Sutron R., Singh S., Predictive Representations of State [C]. Proc. of the Advances in Neural Information Processing Systems, 2002; James M., Wolfe BD, Singh S., Combining Memory and Landmarks with Predictive State Representations [C]. Proceedings of the international Joint Conference on Artificial Intelligence, 2005; proposed and improved a predictive state representation (PSR) dynamic system model. This model uses the probability of observable events directly as the object of statistical analysis, eliminating the performance differences caused by the difficulty in determining the level and number of hidden variables. Wolfe BD, Modeling Dynamical Systems with Structured Predictive State Representations [D], The University of Michigan, 2009, applied it to the design of autonomous driving algorithms with good results. Compared with traditional hidden Markov models, PSR has strong representation capabilities and is intuitive and concise, making it very suitable for describing radar signals.

[0005] Research on Multifunctional Radar Behavior Identification and Prediction Technology [D], 2017, National University of Defense Technology; uses PSR to characterize and analyze signal sequence prediction and operating mode recognition based on radar syntax structure. The authors used a suffix array-based preprocessing algorithm to model the system, using dimensionality reduction and noise reduction to reduce the model size. Using maximum likelihood estimation and finite state automata, they calculated the operating mode probabilities corresponding to radar words and designed a pipeline algorithm from waveform sequence reception and processing to operating mode recognition and estimation. However, this algorithm was unable to recognize mixed-mode waveform sequences, and the training process was complex and inefficient, leaving it somewhat far from practical application. Summary of the Invention

[0006] The embodiments of the present application provide a radar operating mode recognition method and device based on predicted state representation, propose a mapping rule between radar phrases and operating modes, and estimate the mode value of each time step by finding the maximum posterior probability of the mode sequence under the condition of a known state sequence, thereby completing operating mode recognition.

[0007] The present application provides a radar operating mode recognition method based on predicted state representation. A predefined grammatical structure is used to construct a multi-layered grammatical structure of a mixed "radar word"-"radar phrase"-"radar sentence" structure connected as a rule. The multiple operating modes of the radar are converted into each other according to fixed transition probabilities. The radar operating mode recognition method is used to identify the operating mode corresponding to each state based on the radar operating mode at a certain time step and the state sequence of several subsequent time steps, and predict the operating mode at the next time step. The radar operating mode recognition method includes the following steps:

[0008] Pre-use statistical methods, based on length less than l H The state sequence of the state, the mode at time step t is from ω t Change to ω t+1 The probability of constructing a two-dimensional matrix to train the probability model

[0009] Calculate the initial path according to tl H Known state at time step and known patterns , from the probability model Find tl H ~tl H +1 probability of each possible working mode transition And the corresponding mode transfer initial path The mode transfer initial path is used to describe the changes between the working modes of each state in the time step;

[0010] Calculate the sub-paths for each time step, based on the initial path and Find and calculate tl H +1~t each time step in the mode transition probability and its corresponding mode transition path length;

[0011] Calculate the total path and determine tl based on the initial path and the length of each sub-path H The shortest transfer path of the model transfer within the time step t+1 to t+1 And the corresponding path value R t ;

[0012] Voting, move the sliding window backward, repeat the above steps, vote for different recognition results at the same time step according to the corresponding path value, and determine the optimal recognition result based on the voting results.

[0013] Optionally, the following contents may also be predefined:

[0014] Time step τ, the time it takes for the radar to complete a radar word is called a time step;

[0015] The state S at the τth time step τ , at a certain time step, the observable and measurable parameter label emitted by the radar, that is, a radar word represents a state;

[0016] State sequence, a sequence of states in the order of time steps;

[0017] Working mode ω at the τth time step τ ,Each working mode corresponds to a radar phrase, which is generated by selecting one from the set of radar phrases corresponding to the working mode;

[0018] Pattern sequence, a sequence of patterns consisting of time steps.

[0019] Optional, based on the found tl H +1 time step possible mode transition probability, and the corresponding mode transition path, calculate tl H The possible mode transition probability and its corresponding mode transition path length in each time step from +1 to t include:

[0020] Initial τ = tl H +1;

[0021] k=1;

[0022] From the probability model according to the state sequence Finding Mode Transition Probabilities If it can be found, the corresponding mode transfer path length is recorded. If it cannot be found, set k = k + 1 and repeatedly search for the mode transition probability from the probability model according to the state sequence;

[0023] τ=τ+1, if τ≤t, then k=1, and the pattern transition probability is repeatedly searched from the probability model according to the state sequence.

[0024] Optionally, voting is performed based on the determined shortest mode transition paths and corresponding path values, and determining the working mode path according to the voting results includes:

[0025] Record the determined shortest mode transfer paths and corresponding path values. Use the estimated working modes of the shortest mode transfer paths as candidate working modes, and use the corresponding path values ​​as the number of votes.

[0026] If t≤l H , let t=t+1, return to the initial path calculation step S502 and execute again.

[0027] From the obtained H +1 candidate working mode selected path value R tThe minimum path is taken as the working mode recognition result.

[0028] Targeting tl H +1 time step working mode, already has l H +1 "candidate mode". Compare the path value of each estimate Select the working mode with the least number of votes as the final estimate Voting is closed.

[0029] An embodiment of the present application also proposes a radar operating mode recognition device based on predicted state representation, including a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the radar operating mode recognition method based on predicted state representation as described above are implemented.

[0030] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned radar operating mode recognition method based on predicted state representation are implemented.

[0031] The embodiment of the present application proposes to estimate the mode value of each time step by finding the maximum a posteriori probability of the mode sequence under the condition of a known state sequence based on the mapping rule between radar phrases and working modes, thereby completing the identification of the radar working mode.

[0032] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0034] Figure 1 This is a schematic diagram of the basic process of the radar working mode recognition method exemplified in this application;

[0035] Figure 2 This is a schematic diagram of the calculation of the mode transfer path length of the radar working mode recognition method exemplified in this application;

[0036] Figure 3 This is a schematic diagram of the mode transfer path filling method of the radar working mode recognition method exemplified in this application;

[0037] Figure 4 This is a schematic diagram of the specific implementation of the voting method for radar working mode recognition method exemplified in this application;

[0038] Figure 5 The schematic diagram of the "Mercury" radar mode transition probability is an example of the application of the identification method of this application;

[0039] Figure 6 A table showing the working mode and radar phrases of the "Mercury" radar used as an example of the application of the recognition method of this application;

[0040] Figure 7 This is a schematic diagram of the recognition effect of the shortest path method in the application example of this application;

[0041] Figure 8 This is a diagram illustrating the voting method recognition effect of an application example of this application;

[0042] Figure 9 This is an illustration of the recognition effect of the long-short-term memory model used for comparison in this application. DETAILED DESCRIPTION

[0043] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0044] An embodiment of the present application provides a radar operating mode recognition method based on predicted state representation. A grammatical structure is predefined, and radar signal sequences are used as a rule-connected multi-layer grammatical structure of a mixed "radar word"-"radar phrase"-"radar sentence". The multiple operating modes of the radar are converted into each other according to fixed transition probabilities. The radar operating mode recognition method is used to identify the operating mode corresponding to each state based on the radar operating mode at a certain time step and the state sequence of several subsequent time steps, and predict the operating mode of the next time step.

[0045] The following are predefined:

[0046] Time step τ, the time it takes for the radar to complete a radar word is called a time step;

[0047] The state S at the τth time step τ , at a certain time step, the observable and measurable parameter label emitted by the radar, that is, a radar word represents a state, which can also be called an observation.

[0048] State sequence, a sequence of states arranged in the order of time steps, with the subscripts representing the end and start time steps of the sequence respectively.

[0049] Working mode ω at the τth time step τ ,Each working mode corresponds to a radar phrase, which is generated by selecting one from the set of radar phrases corresponding to the working mode;

[0050] Pattern sequence, a sequence of patterns consisting of time steps. The superscripts and subscripts represent the end and start time steps of the sequence respectively.

[0051] The method of this embodiment is to know the state sequence of t time steps Under the condition of t+1, find the mode value ω at time step t+1. t+1 , that is, In order to improve the accuracy of the estimation, we adopt the method of "determining the pattern by sequence" to first find the known state sequence The most likely corresponding pattern sequence And extrapolate 1 time step mode. Mathematically expressed as:

[0052]

[0053] The physical meaning of formula (1) is: find a mode value ω t+1 , so that the state sequence The corresponding pattern sequence The most likely to occur is to maximize the posterior probability.

[0054] If we use the Bayesian formula to calculate the probability of the above formula directly, there will be a problem: due to data sparsity, in most cases it is impossible to directly count the number of long sequences in the existing data. That is to say, due to The sample size is relatively small, and the existing sample size is limited. The event may have never occurred before, so it is difficult to directly estimate its probability.

[0055] The present embodiment complies with the following two assumptions:

[0056] Assumption 1: Markov assumption of pattern sequence. If the current pattern t is in an unfinished state, the next pattern ω t+1 Must be equal to the current mode ω t Otherwise, the next mode will be transferred to another mode according to the probability. t+1 Depends only on the current mode t , and the past pattern sequence Not relevant.

[0057] Assumption 2: Causality. The probability of the pattern at time step t is only regulated by the state sequence before time step t-1, and not by the state sequence at time step t and after. That is:

[0058]

[0059] The two assumptions are in line with common sense and are easy to meet in practice.

[0060] According to Bayes' formula:

[0061]

[0062] Due to assumption 1, the first factor of the above formula

[0063]

[0064] The second factor in formula (2)

[0065]

[0066] The second equation in (4) uses two assumptions at the same time. Substituting (3) and (4) into (2), we can see that it is a recursive equation, so:

[0067]

[0068] because and is a known quantity, so the first factor in the above formula does not affect parameter optimization and can be ignored. To find the maximum value of the above formula, we can take its logarithm and convert the multiplication into addition to form the likelihood function:

[0069]

[0070] In this way, the solution to the maximum a posteriori probability is transformed into a fence-type shortest path problem.

[0071]

[0072] As can be seen from the above steps, the core of this shortest path method lies in finding the mode transition probability under a given state sequence, and then transforming the problem into finding the shortest path and the minimum value of the path. The Viterbi method can effectively solve this type of shortest path problem.

[0073] Specific examples Figure 1 As shown, the radar working mode recognition method proposed in the embodiment of the present application further includes the following steps:

[0074] In step S501, the training model is pre-trained using statistical methods based on the length less than l HThe state sequence of the state, the mode at time step t is from ω t Change to ω t+1 The probability of constructing a two-dimensional matrix to train the probability model In a specific example, the statistical method can be completed using existing mature methods such as the suffix array method.

[0075] Initial path calculation step: In step S502, according to t1 H Known state at time step and known patterns From the probability model Find tl H +1 time step possible working mode transition probability And the corresponding mode transfer initial path The mode transition path is used to describe the changes between the working modes of each state.

[0076] Path calculation step: In step S503, based on the initial transfer path and Find and calculate tl H +1~t each time step and its corresponding mode transition probability and mode transition sub-path length, that is, sub-path length, are as follows: Figure 2 shown.

[0077] Total path calculation step: In step S504, from t1 H +1~t, the mode transition probability that may occur in each time step and its corresponding mode transition path length are determined to determine tl H The shortest mode transfer path within time steps t+1 to t+1 and the corresponding path value R t , specifically Figure 3 shown.

[0078] After determining tl H The shortest mode transfer path within time steps t+1 to t+1 and the corresponding path value R t Afterwards, the shortest mode transition path and the corresponding path value may be stored in step S505. It should be noted that the storage is a specific example, but is not limited thereto.

[0079] Voting step: In step S506, set the sliding window, move the sliding window backward, repeat the above steps, and vote on different recognition results at the same time step according to the corresponding path value. Figure 4 Based on the determined shortest mode transfer paths and corresponding path values, voting is performed, and the working mode path is determined according to the voting results.

[0080] The embodiment of the present application proposes to estimate the mode value of each time step by finding the maximum a posteriori probability of the mode sequence under the condition of a known state sequence based on the mapping rule between radar phrases and working modes, thereby completing the identification of the radar working mode.

[0081] For the sub-path calculation steps, such as Figure 2 As shown, in some embodiments, based on the found tl H +1 time step possible mode transition probability, and the corresponding mode transition path, calculate tl H The possible mode transition probability and its corresponding mode transition path length in each time step from +1 to t include:

[0082] Initial τ = tl H +1;

[0083] k=1;

[0084] From the probability model according to the state sequence Finding Mode Transition Probabilities If it can be found, the corresponding mode transfer path length is recorded. If the query fails, set k=k+1 and repeatedly search for the pattern transition probability from the probability model according to the state sequence.

[0085] τ=τ+1, if τ≤t, then k=1, and the pattern transition probability is repeatedly searched from the probability model according to the state sequence, otherwise the subsequent steps are executed to calculate the shortest path and path value.

[0086] According to the path values ​​determined in the sub-path calculation step, tl is obtained using existing algorithms such as the Viterbi algorithm. H The shortest path within time steps t+1 to t+1 and the corresponding path value R t (like Figure 3 shown).

[0087] In some embodiments, voting is performed based on the determined shortest mode transition paths and corresponding path values, such as Figure 4 As shown, the working mode paths determined according to the voting results include:

[0088] Record the shortest mode transfer paths and corresponding path values ​​determined. and R t They are stored in the estimation mode register and the shortest path register respectively.

[0089] The estimated working modes of each shortest mode transition path are used as candidate working modes, and the corresponding path value is used as the number of votes.

[0090] If t≤lH , let t=t+1, return to the initial path calculation step S502 and execute again.

[0091] From the obtained H +1 candidate working mode selected path value R t The minimum path is taken as the working mode recognition result.

[0092] Targeting tl H +1 time step working mode, already has l H +1 "candidate mode". Compare the path value of each estimate Select the working mode with the least number of votes as the final estimate Voting is closed.

[0093] The radar working mode recognition method of the embodiment of the present application is used to perform a performance test of the pattern recognition algorithm using the "Mercury" radar. The radar working mode transition probability is as follows: Figure 5 As shown, the "Mercury" radar working mode and radar phrase examples are as follows Figure 6 As shown. 200 radar word sequences are randomly generated, and the shortest path method and voting method are used for pattern recognition. The pattern prediction results are used as the starting value of the next time step prediction. The estimation results are shown as follows: Figure 7 、 Figure 8 As shown. As a control, a 200-parameter, 2-layer long short-term memory neural network (LSTM) was trained to recognize the same sequence, and the results were as follows Figure 9 shown.

[0094] Figures 7 to 9 The horizontal axis represents time steps, and the vertical axes 1 to 5 represent the five operating modes: search, interception, non-adaptive tracking, range resolution, and track-hold. Comparison with actual data shows that both the shortest path method and the voting method have strong predictive capabilities. Within the same pattern, both maintain high accuracy. The shortest path method has large estimation errors at the beginning of each pattern, but is relatively accurate in other situations. The voting method overcomes the shortest path method's limitations by optimizing and filtering several estimation results at the same time step, correcting for the bias introduced by the shortest path method when identifying cross-modal states, thereby maintaining overall high prediction accuracy. In contrast, the LSTM method has lower prediction accuracy, particularly in areas with drastic pattern changes. Furthermore, due to the lack of a comprehensive representation of the model structure, LSTM training takes significantly longer.

[0095] The radar working mode recognition method of this application fully considers the information required for cross-mode estimation through the idea of ​​"determining the mode by sequence". Compared with the method mentioned in "Background Art" based on the estimation of the sequence within the same mode, the embodiment of this application establishes an estimation model that conforms to the basic causal logic. As a basic model, it reflects that the pattern of time step t+1 is related to the pattern of time step t and the pattern of length l before time step t. H This design conforms to the basic logic of radar waveform sequence generation and can therefore accurately characterize the relationship between pattern and sequence.

[0096] The shortest path method of this application can not only realize pattern recognition, but also predict patterns. From the perspective of the calculation process, this application proposal can estimate the pattern value of a time step based on the state sequence before the time step. This feature allows the algorithm to not only be used for real-time estimation of pattern sequences, but also to predict the possible pattern value of the next time step, and the algorithm has strong real-time performance. This function can be further applied to a wider range of situations such as radar word prediction, providing the necessary pattern information for dynamic games, pre-interference, smart interference, effect evaluation, etc.

[0097] The voting method proposed in the embodiment of the present application effectively avoids the uncertainty brought about by cross-modal sequence estimation. Since the starting and ending points of each mode in the sequence are not known in advance, the probability of misidentification increases significantly when the sliding window crosses different modes. The voting method closely combines the selection process with the prediction process, and the present application designs an optimization mechanism that uses the shortest path as a criterion. The estimated results during the sliding window change process are stored and screened in a timely manner, and the estimated results when the reference state sequence is within the mode are automatically selected, thereby effectively reducing the negative impact of cross-modal estimation. The voting method is slightly weaker in real time, but has higher accuracy and is more suitable for post-sequence analysis and processing.

[0098] The radar operating mode recognition method proposed in this application effectively avoids the shortcomings of existing algorithms. The algorithm can be directly applied to mixed-mode scenarios without pre-segmenting the state sequences of different modes. The model can be trained directly from a long sequence, eliminating the need to construct fixed-period templates based on the pattern. The algorithm also has strong recognition capabilities for random state templates of the same pattern, and a corresponding smoothing mechanism is set for events that do not appear in the training samples, avoiding a significant increase in errors caused by sample sparsity.

[0099] An embodiment of the present application also proposes a radar operating mode recognition device based on predicted state representation, including a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the radar operating mode recognition method based on predicted state representation as described above are implemented.

[0100] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned radar operating mode recognition method based on predicted state representation are implemented.

[0101] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0102] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0103] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server or network device, etc.) to execute the methods described in each embodiment of the present application.

[0104] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are protected by this application.

Claims

1. A radar operating mode recognition method based on predicted state representation, characterized in that: A predefined grammatical structure is used to construct a radar signal sequence as a multi-layered grammatical structure of a mixed "radar word"-"radar phrase"-"radar sentence" connected by rules. Multiple radar operating modes are converted into each other according to fixed transition probabilities. The radar operating mode recognition method is used to identify the operating mode corresponding to each state based on the radar operating mode at a certain time step and the state sequence of several subsequent time steps, and predict the operating mode at the next time step. The radar operating mode recognition method includes the following steps: Pre-use statistical methods, based on length less than l H The state sequence of the state, the mode at time step t is from ω t Change to ω t+1 The probability of constructing a two-dimensional matrix to train the probability model Calculate the initial path according to tl H Known state at time step and known patterns From the probability model Find tl H ~tl H +1 probability of each possible working mode transition And the corresponding mode transfer initial path The mode transfer initial path is used to describe the changes between the working modes of each state in the time step; Calculate the sub-paths for each time step, based on the initial path and Find and calculate tl H +1~t each time step in the mode transition probability and its corresponding mode transition path length; Calculate the total path and determine tl based on the initial path and the length of each sub-path H The shortest transfer path of the model transfer within the time step t+1 to t+1 And the corresponding path value R t ; Voting, move the sliding window backward, repeat the above steps, vote for different recognition results at the same time step according to the corresponding path value, and determine the optimal recognition result based on the voting results.

2. The radar operating mode recognition method based on predicted state representation according to claim 1, characterized in that: It also includes predefined content: Time step τ, the time it takes for the radar to complete a radar word is called a time step; The state S at the τth time step τ , at a certain time step, the observable and measurable parameter label emitted by the radar, that is, a radar word represents a state; State sequence, a sequence of states in the order of time steps; Working mode ω at the τth time step τ ,Each working mode corresponds to a radar phrase, which is generated by selecting one from the set of radar phrases corresponding to the working mode; Pattern sequence, a sequence of patterns consisting of time steps.

3. The radar operating mode recognition method based on predicted state representation according to claim 2, characterized in that: Based on the found tl H +1 time step possible mode transition probability, and the corresponding mode transition path, calculate tl H The possible mode transition probability and its corresponding mode transition path length in each time step from +1 to t include: Initial τ = tl H +1; k＝1； From the probability model according to the state sequence Finding Mode Transition Probabilities If it can be found, the corresponding mode transfer path length is recorded. If it cannot be found, set k = k + 1 and repeatedly search for the mode transition probability from the probability model according to the state sequence; τ=τ+1, if τ≤t, then k=1, and the pattern transition probability is repeatedly searched from the probability model according to the state sequence.

4. The radar operating mode recognition method based on predicted state representation according to claim 3, characterized in that: Based on the determined shortest mode transfer paths and corresponding path values, voting is performed, and the working mode paths determined according to the voting results include: Recording the determined shortest mode transfer paths and corresponding path values; The working modes of each estimated shortest mode transfer path are used as candidate working modes, and the corresponding path value is used as the number of votes; t=t+1, return to the initial path calculation process, identify until t>l H ; From the obtained H +1 candidate working mode selected path value R t The minimum path is taken as the working mode recognition result.

5. A radar operating mode recognition device based on predicted state representation, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the radar working mode recognition method based on predicted state representation are implemented as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the radar operating mode recognition method based on predicted state representation according to any one of claims 1 to 4.

Images