An interference intention reasoning method, a storage medium and a device
By training recurrent neural networks and hidden Markov models, the intentions of radar jammers can be identified and inferred, solving the problem that radar operators have difficulty accurately obtaining jamming intentions, improving the accuracy of jamming intention inference, and enhancing the radar's anti-jamming capability.
Patent Information
- Application Number
- CN202310278444.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-21
- Publication Date
- 2025-11-04
- Estimated Expiration
- 2043-03-21
AI Technical Summary
In existing technologies, radar systems struggle to accurately obtain information about the jammer's potential intentions, resulting in low accuracy in inferring jamming intentions and an inability to effectively counter jamming in electronic warfare.
By training a recurrent neural network on electronic interference pattern data, the types of interference behaviors are identified, and the interference intent is inferred using the Viterbi algorithm and hidden Markov model. The inference accuracy is improved by combining the correspondence between interference behaviors and intents.
This method achieves high-accuracy reasoning of jamming intentions under uncertain conditions, improves the radar's anti-jamming capability, and provides an effective method for reasoning jamming intentions.
Smart Images
Figure CN116224248B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar anti-jamming technology, and in particular to a method, storage medium, and device for inferring jamming intent. Background Technology
[0002] Radar is indispensable in modern warfare, and jamming methods against it are constantly evolving, leading to an increasingly fierce battle between radar and jammers. In this contest, the jammer holds the initiative, employing various methods and measures to disrupt or even disable the radar's normal target detection capabilities. The radar, on the other hand, is in a relatively passive position, needing to minimize or avoid the impact of jamming. Obtaining as much information as possible about the jammer, such as their behavior and intentions, would greatly enhance the radar's initiative in the confrontation, making its countermeasures against deception jamming more targeted and thus improving its cognitive countermeasure capabilities.
[0003] In jamming countermeasures, jammers need to employ a series of jamming behaviors to achieve a specific jamming intent, specifically by emitting jamming signals of different patterns and modulations. There is a strong correspondence between jamming behaviors and jamming patterns, and jamming behaviors can be identified through jamming pattern recognition results. However, currently, radar operators cannot obtain direct jamming intentions from enemy electronic warfare experts. They can only infer the enemy's hidden intentions through observable changes in jamming behaviors or other observation parameters. Since the same jamming intent may correspond to multiple jamming behaviors, and the current jamming behavior is related to several previous jamming behaviors, there is significant uncertainty between the jamming behavior identified based on perceived information and the subjective intent. Therefore, it is difficult to obtain information about the jammer's potential intentions, leading to low accuracy in inferring jamming intentions. Summary of the Invention
[0004] The purpose of this invention is to address the problem that existing interference intent reasoning methods still have significant uncertainties in the interference behavior and subjective intent identified based on perceived information, which makes it difficult to obtain information about the interference party's potential intent and thus leads to low accuracy in interference intent inference. Therefore, this invention proposes an interference intent reasoning method, storage medium, and device.
[0005] The specific process of a method for inferring intent through interference is as follows:
[0006] Step 1: Obtain the electronic interference pattern data to be detected, preprocess the electronic interference pattern data to be detected, and input the preprocessed electronic interference pattern data to be detected into the electronic interference behavior recognition network to obtain the electronic interference behavior type.
[0007] The electronic jamming pattern dataset includes: aiming jamming, jamming, delayed decoy jamming, frequency-shifting decoy jamming, range dragging jamming, velocity dragging jamming, aiming jamming + dense decoy jamming, intermittent sampling and forwarding jamming, range decoy + velocity decoy jamming, and range dragging + dense decoy jamming.
[0008] Step 2: Input the electronic interference behavior type into the interference intent reasoning model, and use the Viterbi algorithm to obtain the interference intent sequence with the highest probability as the interference intent reasoning result;
[0009] The interference intentions include: reducing detection, affecting confirmation, escaping tracking, and disrupting identification.
[0010] Furthermore, the preprocessing of the electronic interference pattern data to be detected specifically includes:
[0011] Multiple electronic interference pattern data of the same pulse length are spliced together into a single sample. The amplitude of the spliced sample is then used to convert the spliced sample into one-dimensional data, i.e., the pre-processed electronic interference sample data to be detected.
[0012] Furthermore, the electronic interference behavior recognition network is obtained through the following means:
[0013] S1. Obtain the electronic interference pattern dataset, preprocess the electronic interference pattern data, and divide the preprocessed electronic interference pattern dataset into a training set and a test set.
[0014] S2. Construct a recurrent neural network and train the recurrent neural network using the training set until the loss function converges to obtain a trained recurrent neural network.
[0015] S3. Test the trained recurrent neural network using the test set to obtain the recognition accuracy of the trained recurrent neural network. If the recognition accuracy is greater than or equal to the preset threshold, save it as an electronic interference behavior recognition network. If the recognition accuracy is less than the preset threshold, re-execute S1.
[0016] Preferably, the recurrent neural network comprises: an input layer, three hidden layers, and a fully connected layer.
[0017] Furthermore, the interference intent inference model is obtained in the following way:
[0018] Step 1: Obtain the dataset of interfering behavior intentions and divide it into a training set and a test set;
[0019] The dataset of interference behavior intentions includes: interference behavior and interference intention;
[0020] Step 2: Build a Hidden Markov Model (HMM). Train the HMM using the training set to obtain its parameters.
[0021] The EM algorithm is used to estimate the probability of interference intent transfer and the probability of interference behavior output of the Hidden Markov Model using the training set. After the probability values are stable, the parameters of the Hidden Markov Model are determined.
[0022] Step 3: Test the Hidden Markov Model with determined parameters using the test set, and obtain the accuracy of the Hidden Markov Model with determined parameters. If the accuracy is greater than the preset first threshold, then the Hidden Markov Model with determined parameters is the interference intention inference model.
[0023] Furthermore, the interference behaviors include: forming suppression, forming false targets, forming dragging, forming suppression + deception, and forming deception + deception.
[0024] Furthermore, the interference intent includes: reducing detection, affecting confirmation, evading tracking, and disrupting identification.
[0025] A storage medium for interfering intent reasoning, the storage medium storing at least one instruction, the at least one instruction being used in the aforementioned interfering intent reasoning method.
[0026] An interference intent reasoning device, the device comprising: a processor and a memory, the memory storing at least one instruction; the at least one instruction being loaded and executed by the processor to perform the interference intent reasoning method.
[0027] The beneficial effects of this invention are as follows:
[0028] This invention utilizes a recurrent neural network trained on an electronic jamming pattern training set. Under the condition that each jamming behavior corresponds to multiple jamming signal patterns, it can complete a "many-to-many" recognition task, effectively identifying jamming behavior types and providing the necessary conditions for correctly inferring jamming intent. When there is significant uncertainty in the correspondence between jamming intent and jamming behavior, this invention uses the recognition results of a hidden Markov model for ground truth estimation of jamming behavior intent and a recurrent neural network to infer jamming intent. This makes it easier to obtain the jammer's potential intent information, thereby improving the accuracy of jamming intent inference. Embodiments of this invention demonstrate that it has high inference accuracy and provides an effective method for inferring jamming intent for radar anti-jamming. Attached Figure Description
[0029] Figure 1 This is a flowchart of the present invention;
[0030] Figure 2(a) is a time-domain diagram of the intermittent sampling and forwarding interference signal pattern;
[0031] Figure 2(b) shows the range domain diagram of the intermittent sampling and forwarding interference signal pattern;
[0032] Figure 3(a) is a time-domain diagram of the drag interference signal;
[0033] Figure 3(b) shows the range domain diagram of the drag interference signal;
[0034] Figure 4(a) shows a set of true values for the interference behavior;
[0035] Figure 4(b) shows a set of ground truth values for interference intent;
[0036] Figure 5 This is a schematic diagram of a recurrent neural network;
[0037] Figure 6 This is a schematic diagram of a Hidden Markov Model;
[0038] Figure 7 This is a schematic diagram of a interference intent reasoning model;
[0039] Figure 8 Flowchart of the interference intent reasoning method;
[0040] Figure 9 The result of interference behavior identification for recurrent neural networks;
[0041] Figure 10 The inference result is a result of interference with the inference model. Detailed Implementation
[0042] Specific implementation method one: as follows Figure 1 As shown, the specific process of the interference intent reasoning method in this embodiment is as follows:
[0043] Step 1: Obtain the electronic interference pattern data to be detected. Preprocess the electronic interference pattern data and input the preprocessed data into the electronic interference behavior recognition network to obtain the electronic interference behavior type, such as... Figure 9 As shown;
[0044] The preprocessing of the electronic interference pattern data to be detected specifically involves:
[0045] Four electronic interference pattern data of the same pulse length are spliced together into a sample. The amplitude of the spliced sample is used to convert the spliced sample into one-dimensional data, i.e., the pre-processed electronic interference sample data to be detected.
[0046] The electronic interference behavior recognition network is obtained through the following methods:
[0047] S1. Obtain the electronic interference pattern dataset, preprocess the electronic interference pattern dataset, and divide the preprocessed electronic interference pattern dataset into training set and test set;
[0048] The electronic interference pattern data includes multiple interference signal pulses, with a pulse length of 6000;
[0049] As shown in Figures 2(a)-(b), the electronic jamming pattern dataset includes various types of electronic jamming such as aiming jamming, jamming jamming, delay decoy jamming, frequency shift decoy jamming, range dragging jamming, velocity dragging jamming, aiming jamming + dense decoy jamming, intermittent sampling and forwarding jamming, range decoy + velocity decoy jamming, and range dragging + dense decoy jamming. These types of jamming behaviors are respectively classified as suppression, decoy formation, dragging, suppression + deception, and deception + deception, as shown in Table 1.
[0050] Table 1
[0051]
[0052] Because the interference pattern signal contains dragging interference as shown in Figures 3(a)-(b), the network recognition accuracy is related to the number of interference pulses in a sample, i.e., the sample data length. After multiple experiments, interference signals with a length of 4 pulses were spliced together into a single sample as the network input. The interference signal includes real and imaginary data. To facilitate the training and recognition of the recurrent neural network, its amplitude was converted into one-dimensional data and input into the recurrent neural network. The dataset of the interference behavior recognition network contains 2000 sets of samples and corresponding behavior labels. The data is randomly divided into a training set and a test set, with the test set accounting for 50% of the total samples. The training set is mainly used for training the model parameters, and the remaining simulated samples are used as the test set to verify the effectiveness of the trained model. The training and test data diagrams of the interference behavior intent are shown in Figures 4(a)-(b). Each set of data contains 100 rounds of ground truth. The training data is used to estimate the parameters of the Hidden Markov Model, the interference behavior ground truth in the test data is used to generate interference patterns, and the interference intent ground truth is used to test the inference accuracy of the model.
[0053] S2. Construct a recurrent neural network and train the recurrent neural network using the training set until the loss function converges to obtain a trained recurrent neural network.
[0054] The recurrent neural network includes: one input layer, three hidden layers, and one fully connected layer, as follows: Figure 5 As shown.
[0055] The model's learning rate is 0.01, the loss function is cross-entropy loss, and the training algorithm is stochastic gradient descent with backpropagation. The batch size is set to 64. The interference signal includes real and imaginary data; to facilitate training and recognition by the recurrent neural network, its amplitude is converted to one-dimensional input. The parameters of the recurrent neural network model are shown in Table 2.
[0056] Table 2
[0057]
[0058] S3. Test the trained recurrent neural network using the test set to obtain the recognition accuracy of the trained recurrent neural network. If the recognition accuracy is greater than or equal to the preset threshold, save it as an electronic interference behavior recognition network. If the recognition accuracy is less than the preset threshold, re-execute S1.
[0059] Step 2: Input the electronic interference behavior type into the interference intent inference model, and use the Viterbi algorithm to find the interference intent sequence with the highest probability as the interference intent inference result, such as... Figure 8 As shown;
[0060] The interference intent inference model is obtained through the following method:
[0061] Step 1: Obtain the dataset of interfering behavior intentions, and divide the dataset into training set and test set.
[0062] The intent behind the interference behavior includes: the interference behavior and the corresponding interference intent;
[0063] The interference intentions include: reducing detection, affecting confirmation, escaping tracking, and disrupting identification;
[0064] The interference behaviors include: forming suppression, forming false targets, forming dragging, forming suppression + deception, and forming deception + deception;
[0065] Step 2: Based on the interference behavior and interference intent set in the simulation, establish a hidden Markov model of interference behavior and intent. Use the EM algorithm to estimate the interference intent transition probability and interference behavior output probability of the hidden Markov model using the training set. After the probability values are stable, determine the parameters of the hidden Markov model.
[0066] Use the interference process Figure 6 The hidden Markov model shown is modeled as follows: The four intentions of the interfering party—reducing detection, influencing confirmation, escaping tracking, and disrupting identification—are denoted as S1, S2, S3, and S4, respectively. The changes between the interfering intentions can be described by the transition probabilities.
[0067] The radar operator cannot directly know the true intention of the jammer, but it can observe five types of jamming behaviors: suppression, deception, dragging jamming, suppression + deception, and combined deception + deception, denoted as v1, v2, v3, v4, and v5. Changes in jamming behaviors reflect changes in the jammer's intentions and can be described by observation probability values. The sum of the probabilities of all observed jamming behaviors corresponding to each hidden intention is 1. The Hidden Markov Model for the change in jamming intentions is as follows: Figure 7 As shown.
[0068] Step 3: Use the test set to test the Hidden Markov Model after the parameters are determined, and obtain the accuracy of the Hidden Markov Model after the parameters are determined. If the accuracy is greater than the preset first threshold, then the Hidden Markov Model after the parameters are determined is the interference intention inference model.
[0069] The ground truth values of the interference behaviors in the test set are randomly generated into interference pattern data and fed into a recurrent neural network for recognition. The interference behavior recognition result is then output into a hidden Markov model to output the interference intent inference result. The result is compared with the ground truth value to obtain the accuracy of the test.
[0070] The probability of each interference intent is calculated using the Viterbi algorithm, and the sequence of interference intents with the highest probability is selected as the interference intent inference result.
[0071] First, initialize the sequence and calculate the probability of the first round. Then, iterate the probability of possible intentions based on the probability of interference intention transfer and record the optimal interference intention sequence. When terminating, backtrack the optimal interference intention sequence to obtain the intention inference result.
[0072] Specific Implementation Method Two: A storage medium for interfering with intent reasoning, wherein the storage medium stores at least one instruction, and the at least one instruction is used in the aforementioned method for interfering with intent reasoning.
[0073] Specific Implementation Method 3: A device for interfering with intent reasoning, the device comprising: a processor and a memory, the memory storing at least one instruction; the at least one instruction being loaded and executed by the processor to perform the aforementioned method for interfering with intent reasoning.
[0074] Examples: The beneficial effects of the present invention are verified using the following examples:
[0075] To demonstrate the effectiveness of this invention in identifying electronic interference intent, an interference behavior recognition network was trained before identifying 100 rounds of interference pattern data generated from the interference scenario. If the recognition accuracy on the test set exceeded a preset threshold, it was saved as the electronic interference behavior recognition network. A recurrent neural network trained with a pulse count of 4 and an interference-to-noise ratio of 40dB was used to identify 100 rounds of interference signal patterns randomly generated from the ground truth of interference behavior, and the behavior recognition results were output, such as... Figure 9 As shown, the black blocks represent samples that were incorrectly identified, and the identification accuracy rate is 93%.
[0076] The results of the interference behavior identification are input into the Hidden Markov Model to obtain the inference results of the interference intent, such as... Figure 10As shown, the black blocks represent incorrect inference results. The inference accuracy for reducing detection and identifying sabotage is higher than that for influencing confirmation and escaping tracking intent. This is because the interference behavior of the interfering party when intending to reduce detection and identify sabotage is relatively predictable; for example, when intending to reduce detection, the interfering party often chooses to use suppression interference. From the above experimental results, it can be seen that under the set interference parameters, the inference accuracy of the interference intent inference model based on the Hidden Markov Model is approximately 80%.
Claims
1. An interference intent reasoning method, characterized by The method specifically comprises the following steps: Step 1: obtaining electronic interference pattern data to be detected, preprocessing the electronic interference pattern data to be detected, inputting the preprocessed electronic interference pattern data to be detected into an electronic interference behavior recognition network to obtain an electronic interference behavior type; The electronic interference pattern data set comprises: aiming interference, blocking interference, delay false target interference, frequency shift false target interference, distance drag interference, speed drag interference, aiming interference + dense false target interference, intermittent sampling forwarding interference, distance false target + speed false target interference, distance drag + dense false target interference; Step 2: inputting the electronic interference behavior type into an interference intention reasoning model, and obtaining a probability maximum interference intention sequence as an interference intention reasoning result by using a Viterbi algorithm; The interference intention comprises: reducing detection, affecting confirmation, escaping tracking, and destroying identification.
2. The method of inferring intent according to claim 1, wherein: The preprocessing of the electronic interference pattern data to be detected comprises the following steps: Splicing a plurality of electronic interference pattern data to be detected of the same pulse length into one sample, and converting the sample into one-dimensional data by using the amplitude of the sample, thereby obtaining the preprocessed electronic interference pattern data to be detected.
3. The method of inferring intent according to claim 2, wherein: The electronic interference behavior recognition network is obtained by the following method: S1: obtaining an electronic interference pattern data set, preprocessing the electronic interference pattern data, and dividing the preprocessed electronic interference pattern data set into a training set and a test set; S2: constructing a recurrent neural network, training the recurrent neural network by using the training set until the loss function converges to obtain a trained recurrent neural network; S3: testing the trained recurrent neural network by using the test set, obtaining the recognition accuracy of the trained recurrent neural network, and saving the trained recurrent neural network as an electronic interference behavior recognition network if the recognition accuracy is greater than or equal to a preset threshold, or re-executing S1 if the recognition accuracy is less than the preset threshold.
4. The method of inferring intent according to claim 3, wherein: The loss function of the recurrent neural network is a cross-entropy loss function.
5. The method of inferring intent according to claim 4, wherein: The recurrent neural network comprises: one input layer, three hidden layers, and one full connection layer.
6. The method of inferring intent according to any one of claims 1-5, wherein: The interference intention reasoning model is obtained by the following method: Step 1: obtaining an interference behavior intention data set, and dividing the interference behavior intention data set into a training set and a test set; The interference behavior intention data set comprises: interference behavior and interference intention; Step 2: establishing a hidden Markov model, training the hidden Markov model by using the training set, and obtaining parameters of the hidden Markov model: Using an EM algorithm to estimate the interference intention transition probability and the interference behavior output probability of the hidden Markov model by using the training set, and determining the parameters of the hidden Markov model after the probability value is stable; Step 3: testing the hidden Markov model with the determined parameters by using the test set, obtaining the accuracy of the hidden Markov model with the determined parameters, and taking the current hidden Markov model with the determined parameters as the interference intention reasoning model if the accuracy is greater than a preset first threshold.
7. The method of inferring intent according to claim 6, wherein: The interference behavior comprises: forming suppression, forming false target, forming drag, forming suppression + deception, and forming deception + deception.
8. An inference intent interfering storage medium characterized by: The storage medium stores at least one instruction for implementing the interference intention reasoning method in any one of claims 1-7.
9. An interference intent reasoning device, comprising: The device comprises a processor and a memory, and the memory stores at least one instruction; the at least one instruction is loaded and executed by the processor to implement the interference intention reasoning method in any one of claims 1-7.
Citation Information
Patent Citations
Random simulation optimization-based range gate pull-off interference method
CN114722710A
Intention recognition method based on bidirectional gating circulation unit and conditional random field
CN114818853A