A non-ideal observation radar PRI modulation identification method

Abstract

The application discloses a non-ideal observation radar PRI modulation identification method, comprising the following steps: presetting parameter ranges of different PRI modulation types; generating a plurality of PRI pulse sequences for each PRI modulation type; preprocessing the PRI pulse sequences generated for each PRI modulation type, generating GASF coded images and GADF coded images, and establishing an identification dataset; establishing a PRI modulation identification model, fusing features output by two convolutional neural networks of the PRI modulation identification model to obtain a network identification result; for a PRI pulse sequence corresponding to each sample, determining a continuity identification result of the PRI modulation type by calculating a continuity parameter of the PRI pulse sequence and based on a prior parameter interval; combining the network identification result and the continuity identification result, and if the continuity identification result contains the network identification result, taking the network identification result as a final identification result; and acquiring a PRI pulse sequence to be identified, determining a PRI modulation type by using the trained PRI modulation identification model and combining a calculation result of a continuity parameter of the PRI pulse sequence.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radar electronic reconnaissance, and in particular to a non-ideal observation radar PRI modulation identification method. BACKGROUND

[0002] Pulse Repetition Interval (PRI) is the interval of the time of arrival (TOA) of adjacent pulses emitted by a radar, and the reciprocal of the pulse repetition frequency. In order to improve the performance and counter-detection capability of the radar, different radar PRI modulation types have emerged. Since PRI has many learnable fixed patterns, it is directly and closely related to the working performance and system of the radar, and thus the identification of the PRI modulation type of the radar emitter is conducive to understanding the working mode and task of the radar emitter. However, with the diversity of PRI modulation modes and the interference of factors such as pulse loss, false pulses, and external noise caused by non-ideal observation of the radar emitter, the accurate identification of the PRI modulation type still faces great challenges.

[0003] PRI modulation type identification methods are mainly divided into two categories. One category performs statistical processing on the TOA sequence, such as extracting its first-order difference features, and then determining the PRI modulation type according to the distribution of the histogram using artificial experience. This category of method relies on artificial features and cannot identify complex PRI modulation modes. The other category of method uses artificial intelligence technology to extract time-domain TOA features and classify them, such as PRI modulation identification methods based on convolutional neural networks and long short-term memory networks. This category of method uses deep learning methods to mine the periodic features between PRI sequences, and has a high identification accuracy rate when the pulse detection is discontinuous and the signal-to-noise ratio is low.

[0004] Since the deep learning method introduces a feature transformation process, it will lose part of the time-domain information in the original pulse sequence, so that the algorithm cannot fully mine the timing rules of the PRI pulse sequence. In the high pulse loss and noise pulse scenario, the identification accuracy rate is not high. Therefore, there is an urgent need for a new PRI modulation type identification method that can ensure that the regularity of the original feature sequence of PRI is not destroyed under non-ideal observation conditions. SUMMARY

[0005] The purpose of the present application is to provide a non-ideal observation radar PRI modulation identification method to overcome the problem of low identification accuracy rate in the prior art.

[0006] In order to achieve the above-mentioned task, the present application adopts the following technical solutions:

[0007] A non-ideal observation radar PRI modulation identification method comprises:

[0008] preset parameter ranges of different PRI modulation types; generate a plurality of PRI pulse sequences for each PRI modulation type;

[0009] generate PRI pulse sequences for each PRI modulation type, and generate GASF encoded images and GADF encoded images using the preprocessed PRI pulse sequences; and establish a recognition dataset using the GASF encoded images and the GADF encoded images;

[0010] construct a PRI modulation recognition model; the recognition model adopts a parallel convolutional neural network, each convolutional neural network of which is used for feature extraction of the GASF encoded images and the GADF encoded images in the sample; and the features output by the two convolutional neural networks are fused to obtain a network recognition result;

[0011] for each PRI pulse sequence corresponding to each sample, determine a continuity recognition result of the PRI modulation type by calculating the continuity parameter thereof and based on the prior parameter interval; and combine the network recognition result and the continuity recognition result, and if the continuity recognition result contains the network recognition result, take the network recognition result as the final recognition result;

[0012] obtain a PRI pulse sequence to be recognized, and determine the PRI modulation type using the trained PRI modulation recognition model and in combination with the calculation result of the continuity parameter thereof.

[0013] simulate according to the parameter range of each PRI modulation type to obtain a plurality of PRI pulse sequences.

[0014] Further, the process of preprocessing the PRI pulse sequence is as follows:

[0015] normalize each PRI pulse in the PRI pulse sequence to obtain a normalized PRI pulse sequence;

[0016] convert the normalized PRI pulse sequence into a polar coordinate system; wherein each normalized PRI pulse is converted into an angle in the polar coordinate system

[0017]

[0018] wherein, represents the i-th PRI pulse in the normalized PRI pulse sequence, and N is the length of the PRI pulse sequence.

[0019] Further, the GASF encoded images and the GADF encoded images are generated using the preprocessed PRI pulse sequence, and are represented as:

[0020]

[0021]

[0022] wherein G GASF , G GADF are the generated GASF encoded image and GADF encoded image respectively; denotes the n-th PRI pulse in the normalized PRI pulse sequence corresponding angle in polar coordinate system.

[0023] Further, for each PRI modulation type, a set of GASF and GADF encoded images is generated for each PRI pulse sequence; the set of GASF and GADF encoded images is taken as a sample in the dataset, and the label of the sample is the PRI modulation type.

[0024] Further, the parallel convolutional neural networks have the same network structure, and each is a CNN network branch; each CNN network branch has a 9-layer network structure, the first layer, the third layer, the fifth layer to the seventh layer are convolutional layers, the second layer, the fourth layer and the eighth layer are pooling layers, and the ninth layer is a fully connected layer; wherein a ReLU activation function is used after each convolutional layer to introduce nonlinearity;

[0025] The features of the GASF encoded image and the GADF encoded image are extracted by the two CNN network branches respectively, then the flattened features are fused in the feature fusion layer, and finally the fused features are input into another fully connected layer to realize the classification and recognition of the sample, and the network recognition result of the PRI modulation type is obtained.

[0026] Further, the continuity parameter determination process of the PRI pulse sequence is as follows:

[0027] Let the continuity threshold of each PRI pulse sequence be:

[0028]

[0029] wherein x i denotes the i-th PRI pulse in the PRI pulse sequence, and the length of the PRI pulse sequence is N;

[0030] Define the continuity parameter C as the ratio of the number of PRI pulses M satisfying continuity to the total length N of the PRI pulse sequence; the number of PRI pulses M satisfying continuity is the number of pulses whose x i and x i-1 , x i+1 difference absolute value is lower than the continuity threshold ε.

[0031] Further, the prior parameter interval of each PRI modulation type is determined in advance by statistical calculation, and then the continuity parameter of each PRI pulse sequence is calculated, and when the continuity parameter falls within the prior parameter interval of a PRI modulation type, the PRI pulse sequence X is considered to be the PRI modulation type corresponding to the interval, and is taken as the continuity recognition result.

[0032] Further, the PRI modulation type includes fixed PRI, jitter PRI, staggered PRI, group variable PRI, sliding PRI, and sinusoidal PRI.

[0033] The sinusoidal PRI parameter range is 100-3000us, the staggered number is 5-20, the sliding step is 10-100us, and the period number is 2-20; the group variable PRI parameter range is 100-3000us, the staggered number is 3-10, and the period number is 10-20; the sliding PRI parameter range is 300-500us, the staggered number is 3-20, the sliding step is 10-100us, and the jitter PRI parameter range is 100-3000us, the jitter range is 2-20%; the staggered PRI parameter range is 100-3000us, the staggered number is 3-20; and the fixed PRI parameter range is 100-3000us.

[0034] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, the non-ideal observation radar PRI modulation recognition method is implemented.

[0035] A computer readable storage medium, the medium stores a computer program; when the computer program is executed by a processor, the non-ideal observation radar PRI modulation recognition method is implemented.

[0036] Compared with the prior art, the present application has the following technical features:

[0037] 1. One-dimensional PRI sequence is converted from Cartesian coordinate system to polar coordinate, Gram matrix is encoded based on trigonometric function, two-dimensional GAF image based on PRI is generated, time correlation is obtained, and hidden features in PRI sequence data are revealed.

[0038] 2. GASF and GADF encoded images are used jointly, and parallel CNN network is used to extract feature information and make fusion, so that multi-dimensional features of PRI modulation type are fully extracted.

[0039] 3. The method can still have relatively high recognition accuracy under the condition of non-ideal observation condition, pulse loss and stray pulse interference. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The sequence diagram of six PRI modulation types in ideal environment, (a) is fixed PRI pulse sequence, (b) is group variable PRI pulse sequence, (c) is staggered PRI pulse sequence, (d) is sinusoidal PRI pulse sequence, (e) is sliding PRI pulse sequence, (f) is jittered PRI pulse sequence;

[0041] Figure 2 The image of six PRI modulation type sequences converted into GASF, (a) is fixed PRI pulse sequence, (b) is group variable PRI pulse sequence, (c) is staggered PRI pulse sequence, (d) is sinusoidal PRI pulse sequence, (e) is sliding PRI pulse sequence, (f) is jittered PRI pulse sequence;

[0042] Figure 3 The image of six PRI modulation type sequences converted into GADF, (a) is fixed PRI pulse sequence, (b) is group variable PRI pulse sequence, (c) is staggered PRI pulse sequence, (d) is sinusoidal PRI pulse sequence, (e) is sliding PRI pulse sequence, (f) is jittered PRI pulse sequence;

[0043] Figure 4 The block diagram of non-ideal observation radar PRI modulation recognition method based on Gram angular field joint coding;

[0044] Figure 5 The structure diagram of parallel CNN network;

[0045] Figure 6 The recognition accuracy under different pulse loss rates and different pulse noise rates, (a) is the recognition accuracy under different loss conditions, (b) is the recognition accuracy under different noise conditions;

[0046] Figure 7 The confusion matrix under different pulse loss rates, (a) is the confusion matrix under 40% loss rate, (b) is the confusion matrix under 50% loss rate;

[0047] Figure 8 The confusion matrix under different pulse noise rates, (a) is the confusion matrix under 40% loss rate, (b) is the confusion matrix under 50% loss rate. DETAILED DESCRIPTION

[0048] This invention addresses the problem of PRI modulation type identification for radar radiation sources by providing a method for identifying PRI modulation in non-ideal observation radar. By transforming the one-dimensional PRI sequence to polar coordinates and utilizing a joint encoding method of trigonometric function differences and sums, the temporal correlation features of different PRI modulation types are extracted. This method achieves high-accuracy identification of six PRI modulation types under non-ideal observation conditions such as pulse loss and pulse interference. See the attached figures for the specific details of the method:

[0049] Step 1: Preset the parameter range for different PRI modulation types; generate multiple PRI pulse sequences for each PRI modulation type.

[0050] In radar applications, the common PRI modulation types are fixed PRI, dithered PRI, staggered PRI, group-variable PRI, slip-variable PRI, and sinusoidal PRI. The parameter range settings for the PRI modulation types in this invention are as follows:

[0051] The parameters for the sinusoidal PRI range from 100 to 3000 μs, with 5 to 20 staggers, a slip step size of 10 to 100 μs, and 2 to 20 cycles; the parameters for the grouped PRI range from 100 to 3000 μs, with 3 to 10 staggers and 10 to 20 cycles; the parameters for the slipped PRI range from 300 to 500 μs, with 3 to 20 staggers and a slip step size of 10 to 100 μs; the parameters for the jittering PRI range from 100 to 3000 μs, with a jitter range of 2% to 20%; the parameters for the staggered PRI range from 100 to 3000 μs, with 3 to 20 staggers; and the parameters for the fixed PRI range from 100 to 3000 μs.

[0052] Table 1 PRI Parameter Settings

[0053]

[0054]

[0055] Based on the parameter ranges of each PRI modulation type mentioned above, simulations were performed to obtain multiple PRI pulse sequences for six PRI modulation types: fixed, jittery, staggered, grouped, slip, and sinusoidal. A single PRI pulse sequence for each modulation type is denoted as X = {x1, x2, ..., x...}. N}, where x i Let represent the i-th PRI pulse, characterized by TOA; the length of the PRI pulse sequence is N.

[0056] Step 2, preprocessing the PRI pulse sequence generated for each PRI modulation type, generating GASF encoded images and GADF encoded images using the preprocessed PRI pulse sequence; using the GASF encoded images and the GADF encoded images to establish a recognition data set.

[0057] Gramian Angle Field (GAF) is a method for feature encoding of one-dimensional time series. The method converts signals in Cartesian coordinate system to polar coordinate system for representation, and then obtains information representing time correlation through the difference and summation of trigonometric functions according to the definition form of inner product, to generate two types of Gramian Angular Difference Field (GADF) and Gramian Angular Summation Field (GASF).

[0058] The specific steps of converting the PRI pulse sequence into GADF and GASF encoded images are as follows:

[0059] Step 21, in order to make the inner product of the input sequence not biased to the maximum observation value, each PRI pulse sequence X = {x1, x2, …, xn} is scaled into the interval [-1, 1], and the formula is as follows: N} into the interval [-1, 1], and the formula is as follows:

[0060]

[0061] wherein, represents the i-th PRI pulse x i The normalized result, max(X), min(X) is the maximum and minimum value in the PRI pulse sequence X.

[0062] Step 22, converting the normalized PRI pulse sequence into polar coordinate system; wherein each normalized PRI pulse is converted into the angle

[0063]

[0064] Step 23, generating GASF and GADF encoded images, and the transformation formulas of the two encoding methods are as follows:

[0065]

[0066] For each PRI modulation type, each PRI pulse sequence can generate a set of GASF and GADF encoded images; a set of GASF and GADF encoded images is taken as a sample in the data set, and the label of the sample is the PRI modulation type.

[0067] Using the same method, samples of different PRI modulation types are generated.

[0068] In an embodiment of the present application, the length of each PRI pulse sequence N=300, 2200 samples are generated for each PRI modulation type, and the data set is divided into a training set, a test set and a validation set according to a ratio of 7:2:1.

[0069] Step 3, constructing a PRI modulation recognition model; the recognition model adopts a parallel convolutional neural network, each of which is used for feature extraction of GASF encoded images and GADF encoded images in the sample; the features output by the two convolutional neural networks are fused to obtain a network recognition result.

[0070] Among them, the parallel convolutional neural network has the same network structure, which is a CNN network branch, one branch processes GASF encoded images, and the other branch processes GADF encoded images.

[0071] Each CNN network branch has a 9-layer network structure, the first layer, the third layer, the fifth layer to the seventh layer are convolutional layers, the second layer, the fourth layer and the eighth layer are pooling layers, and the ninth layer is a fully connected layer; wherein a ReLU activation function is introduced after each convolutional layer to introduce nonlinearity.

[0072] The first layer is a convolutional layer, the convolution kernel size is 7x7, the input channel number is 3, the output channel number is 48, the convolution operation sliding step is 2, and the primary feature extraction is performed, and the global features of the encoded image are captured through a large convolution kernel;

[0073] The second layer is a pooling layer, which adopts a 2x2 window maximum pooling to downsample the input features, thereby reducing the spatial dimension and computational complexity;

[0074] The third layer is a convolutional layer, the convolution kernel size is 5x5, the input channel number is 48, the output channel number is 128, the convolution operation sliding step is 1, and the middle-level feature extraction is performed, and the channel number is expanded to enrich the feature representation;

[0075] The fourth layer is a pooling layer, which adopts a 2x2 window maximum pooling to downsample again, thereby compressing the spatial dimension;

[0076] The fifth layer to the seventh layer all adopt a convolutional layer, the convolution kernel size is 3x3, the feature is refined through high-level feature extraction, a small convolution kernel is used to refine the feature, and the channel number is compressed through continuous convolution, thereby realizing feature refinement;

[0077] The eighth layer is a pooling layer, which adopts a 2x2 window maximum pooling to downsample the input features.

[0078] The ninth layer is a full connection layer, which flattens the features output by the pooling layer.

[0079] The features of the GASF encoded image and the GADF encoded image are extracted through two CNN network branches respectively, then the flattened features are fused in the feature fusion layer, and finally the fused features are input into another full connection layer to realize the classification and recognition of the samples, and the network recognition result of the PRI modulation type is obtained.

[0080] During the training of the PRI modulation recognition model, the loss function adopts cross-entropy error loss, the minibatch size is 32, the optimization algorithm adopts SGD, the learning rate is set to 0.001, and the momentum is set to 0.9; through continuous training and parameter adjustment, until the network performance reaches the best, the current best network model is saved.

[0081] Step 4, for each sample corresponding to the PRI pulse sequence X, the continuity parameter thereof is calculated, and based on the prior parameter interval, the continuity recognition result of the PRI modulation type is determined;

[0082] Combining the network recognition result and the continuity recognition result, if the network recognition result is contained in the continuity recognition result, the network recognition result is taken as the final recognition result.

[0083] The continuity parameter determination process of the PRI pulse sequence X is as follows:

[0084] The i-th PRI pulse x i of the PRI pulse sequence X i The condition for continuity is that the absolute value of the difference between x i-1 , x i+1 is lower than the continuity threshold ε; considering the influence of random jitter, the continuity threshold of each PRI pulse sequence X is set as:

[0085]

[0086] The continuity parameter C is defined as the ratio of the number of PRI pulses satisfying continuity M to the total length N of the PRI pulse sequence X, that is:

[0087] C=M / N(6)

[0088] Wherein, the number of PRI pulses satisfying continuity M is the number of pulses whose absolute value of difference is lower than the continuity threshold ε, and is defined as: i i-1 i+1

[0089]

[0090] f(·) is an indicator function, and is defined as: ​​​

[0091]

[0092] The prior parameter interval of each PRI modulation type is determined in advance by statistical calculation, and then the continuity parameter of each PRI pulse sequence X is calculated, and when the continuity parameter falls into the prior parameter interval of a PRI modulation type, the PRI pulse sequence X is considered to be the PRI modulation type corresponding to the interval, and is taken as the continuity recognition result.

[0093] Therefore, if the network recognition result is included in the continuity recognition result of a certain PRI pulse sequence, the network recognition result is taken as the final recognition result.

[0094] In an embodiment of the present application, the prior parameter interval of the continuity parameter of fixed PRI, sinusoidal PRI and group-variable PRI is 0.8-1, the prior parameter interval of the continuity parameter of staggered PRI and sliding PRI is 0-0.05, and the prior parameter interval of the continuity parameter of jittered PRI is 0.1-0.3.

[0095] That is, for example, the network recognition result of a certain PRI pulse sequence is "fixed PRI", and the calculated value of the continuity parameter C is 0.9, which falls into the prior parameter interval 0.8-1, and the continuity recognition result can be one of "fixed PRI, sinusoidal PRI and group-variable PRI", which includes "fixed PRI", and therefore "fixed PRI" is taken as the final recognition result.

[0096] Step 5: obtaining a PRI pulse sequence to be recognized, determining the network recognition result thereof by using the PRI modulation recognition model trained, and determining the continuity recognition result thereof by using the continuity parameter; if the network recognition result is included in the continuity recognition result, the network recognition result is taken as the final recognition result; otherwise, the next PRI pulse sequence is recognized.

[0097] The PRI modulation type recognition accuracy of the method of the present application under the conditions of PRI pulse sequence loss rate R loss =[10%, 20%, 30%, 40%, 50%] and pulse interference R spurious =[10%, 20%, 30%, 40%, 50%] is verified, and the recognition results are shown in (a) and (b) of Figure 6 With the change of the loss rate, the recognition accuracy of each category changes, and the overall accuracy decreases with the increase of the loss rate and the noise rate.

[0098] It can be seen from the simulation results that the PRI modulation type is less affected by the lost pulses than the noise pulses, and when the loss rate reaches 50%, the PRI type identification accuracy is still above 94%, and when the noise rate reaches 50%, the PRI type identification accuracy is still above 86%. When the radiation source pulse loss rate and noise rate are too large, the number of real pulses in the PRI sequence is too small to reflect the periodicity of the PRI modulation type, and therefore the identification accuracy decreases; Figure 7 and Figure 8 are the identification confusion matrices under different loss rates and noise rates, respectively.

[0099] The above examples are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A non-ideal observation radar PRI modulation identification method, characterized in that, The method comprises the following steps: presetting parameter ranges of different PRI modulation types; generating a plurality of PRI pulse sequences for each PRI modulation type; preprocessing the PRI pulse sequences generated for each PRI modulation type, generating GASF coded images and GADF coded images by using the preprocessed PRI pulse sequences, and establishing an identification dataset by using the GASF coded images and the GADF coded images; constructing a PRI modulation identification model; the identification model adopts parallel convolutional neural networks, each of which is used for feature extraction of the GASF coded images and the GADF coded images in the sample; the features output by the two convolutional neural networks are fused to obtain a network identification result; for each PRI pulse sequence corresponding to a sample, the continuity of the PRI modulation type is determined by calculating the continuity parameter and based on the prior parameter interval; the network identification result is taken as the final identification result if the network identification result is contained in the continuity identification result; obtaining a PRI pulse sequence to be identified, determining the PRI modulation type by using the trained PRI modulation identification model and combining the calculation result of the continuity parameter; simulating according to the parameter range of each PRI modulation type to obtain a plurality of PRI pulse sequences; The continuity parameter determination process of the PRI pulse sequence is as follows: the continuity threshold of each PRI pulse sequence is: wherein, represents the i-th PRI pulse in a PRI pulse sequence, the length of the PRI pulse sequence being ; and ; and The continuity parameter C is defined as the number of PRI pulses satisfying continuity The ratio of the total length of the PRI pulse sequence The number of PRI pulses satisfying continuity The number of PRI pulses satisfying continuity The number of PRI pulses satisfying continuity , The number of pulses whose difference absolute value is lower than the continuity threshold The number of pulses whose difference absolute value is lower than the continuity threshold the prior parameter interval of each PRI modulation type is determined in advance by statistical calculation, so that when the continuity parameter of each PRI pulse sequence falls into the prior parameter interval of a PRI modulation type, the PRI pulse sequence is considered to be the PRI modulation type corresponding to the interval, and is taken as the continuity identification result.

2. The non-ideal observation radar PRI modulation identification method according to claim 1, characterized in that, The preprocessing process of the PRI pulse sequence is as follows: normalize each PRI pulse in the PRI pulse sequence to obtain a normalized PRI pulse sequence; convert the normalized PRI pulse sequence into a polar coordinate system; where each normalized PRI pulse is converted to an angle in polar coordinates : wherein, denotes the i-th PRI pulse in the normalized sequence of PRI pulses, N is the length of the sequence of PRI pulses.​ 3. The non-ideal observation radar PRI modulation identification method according to claim 1, characterized in that, generate GASF coded images and GADF coded images by using the preprocessed PRI pulse sequence, which is represented as: wherein, , are the generated GASF encoded image and GADF encoded image, respectively; denotes the normalized PRI pulse sequence, and denotes the corresponding angle in polar coordinates.

4. The non-ideal observation radar PRI modulation identification method according to claim 1, characterized in that, for each PRI modulation type, a set of GASF and GADF coded images is generated for each PRI pulse sequence; a set of GASF and GADF coded images is taken as a sample in the dataset, and the label of the sample is the PRI modulation type.

5. The non-ideal observation radar PRI modulation identification method according to claim 1, characterized in that, The parallel convolutional neural networks have the same network structure, which are CNN network branches; each CNN network branch has a 9-layer network structure, the first layer, the third layer, the fifth layer to the seventh layer are convolutional layers, the second layer, the fourth layer and the eighth layer are pooling layers, and the ninth layer is a fully connected layer; wherein, a ReLU activation function is introduced after each convolutional layer to introduce nonlinearity. The features of the GASF coded image and the GADF coded image are extracted through two CNN network branches respectively, then the flattened features are fused in a feature fusion layer, finally the fused features are input into another fully connected layer to realize the classification and recognition of the samples, and the network recognition result of the PRI modulation type is obtained.

6. The non-ideal observation radar PRI modulation identification method according to claim 1, characterized in that, The PRI modulation type includes fixed PRI, dithering PRI, staggered PRI, group variable PRI, sliding PRI and sinusoidal PRI. The sinusoidal PRI parameter range is 100-3000us, the staggered number is 5-20, and the period number is 2-20; the group variable PRI parameter range is 100-3000us, the staggered number is 3-10, and the period number is 10-20; the sliding PRI parameter range is 300-500us, the staggered number is 3-20; the dithering PRI parameter range is 100-3000us, the dithering range is 2-20%; the staggered PRI parameter range is 100-3000us, the staggered number is 3-20; and the fixed PRI parameter range is 100-3000us.

7. A terminal device comprising a processor, a memory, and a computer program stored in the memory; characterized in that, The processor executes the computer program to realize the non-ideal observation radar PRI modulation identification method in any one of claims 1-6.

8. A computer readable storage medium having stored therein a computer program; characterized in that, The computer program is executed by the processor to realize the non-ideal observation radar PRI modulation identification method in any one of claims 1-6.

Images