Electromagnetic signal flow clustering method, device, medium and system based on neural network signal high-order characteristics

Abstract

The invention discloses an electromagnetic signal flow clustering method, device, medium and system based on neural network signal high-order characteristics, and belongs to the field of intelligent signal processing. And performing cosine similarity of the signal high-order feature and the existing clustering high-order feature on the nth signal. And if the maximum cosine similarity is greater than a threshold, classifying the signal into a corresponding cluster. Otherwise, creating a new category until all the signals are traversed, thereby realizing electromagnetic signal clustering based on neural network signal high-order features. The method does not depend on electromagnetic parameters of a radiation source, and the problem of electromagnetic signal clustering errors caused by low dimensionality of traditional PDW-based parameters is avoided.

Description

Technical Field

[0001] The present invention relates to the field of intelligent signal processing, and more specifically, to an electromagnetic signal flow clustering method, device, medium and system based on high-order characteristics of neural network signals. Background Art

[0002] Traditional electromagnetic signal clustering is mainly based on manually extracted PDW features, but due to the low dimension of PDW parameters, there is serious parameter overlap, resulting in poor adaptability. The academic community has proposed the extraction of intermediate frequency high-order features of electromagnetic signals based on neural networks, but it is currently only used for target recognition, and there is little research on signal clustering. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an electromagnetic signal stream clustering method, device, medium and system based on the high-order characteristics of neural network signals, which is independent of the electromagnetic parameters of the radiation source and avoids the problem of electromagnetic signal clustering errors caused by the low dimension of traditional PDW parameters.

[0004] The object of the present invention is achieved through the following solutions:

[0005] A method for clustering electromagnetic signal streams based on high-order features of neural network signals comprises the following steps:

[0006] Suppose for the nth signal, the high-order feature of the signal extracted by the neural network feature is represented as Feature n , where n = 1, 2, ..., N, N represents the total number of signals, Feature n The dimension is K-dimensional; for the mth cluster, its clustering high-order features are expressed as Where m = 1, 2, ..., M, M represents the total number of signals, Feature n and The dimensions are all K-dimensional; the first signal is taken as the first cluster, and its clustering high-order features are After initialization is completed, follow these steps:

[0007] S1: For the nth signal, assuming there are m clustering results, the cosine similarity between the nth signal and the mth clustering result is calculated in sequence, which is:

[0008]

[0009] in, Represents the absolute value of the dot product of the nth signal high-order feature and the mth cluster feature, ||Feature n || represents the Euclidean norm of the nth signal’s high-order features, Represents the Euclidean norm of the high-order features of the m-th cluster;

[0010] S2: Find R n,m The maximum value is MaxR n,m , and its corresponding serial number is max_m; set the threshold value to Thres. If MaxR n,m ≥Thres, then classify the n-th signal into the max_m-th cluster and proceed according to S3; if MaxR n,m <Thres, then it is considered that the n-th signal belongs to a new cluster and proceed according to S4;

[0011] S3: For the max_m-th cluster, update its high-order cluster features with the high-order features of the n-th signal,

[0012] That is where α is the smoothing coefficient of the high-order cluster features. When the dot product of the high-order features of the n-th signal and the features of the m-th cluster is greater than or equal to 0, Otherwise

[0013] S4: Add a new cluster, that is, m = m + 1, and set the high-order features of this cluster to the high-order features of the n-th signal, that is

[0014] S5: Starting from the second signal, process according to S1 to S4 in turn until all signals are processed, that is, complete the signal clustering process based on the high-order features of the neural network signals.

[0015] Furthermore, the value of dimension K includes 32.

[0016] Furthermore, the value of the threshold Thres includes 0.6.

[0017] Furthermore, the value of the smoothing coefficient α of the high-order cluster features includes 1.

[0018] An electromagnetic signal flow clustering device based on the high-order features of neural network signals, including a processor and a memory. A computer program is stored in the memory, and when the computer program is loaded by the processor, it executes the method described in any one of the above.

[0019] A computer-readable storage medium stores a computer program, and when the computer program is loaded by a processor, it executes the method described in any one of the above.

[0020] An electromagnetic signal flow clustering system based on the high-order features of neural network signals includes the device described above.

[0021] The beneficial effects of the present invention include:

[0022] The present invention does not rely on the electromagnetic parameters of the radiation source, thus avoiding the problem of electromagnetic signal clustering errors caused by the low dimension of the traditional PDW parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 A flowchart of the method steps of an embodiment of the present invention;

[0025] Figure 2 4 is a block diagram of high-order feature extraction of neural network signals according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0027] In view of the problems in the background, the present invention provides an electromagnetic signal stream clustering method based on high-order features of neural networks. First, the clustering is initialized with the high-order features of the first signal, and then the cosine similarity of the signal high-order features and the existing clustering high-order features is performed for the nth signal. If the maximum cosine similarity is greater than the threshold, the signal is assigned to the corresponding cluster. Otherwise, a new category is created until all signals are traversed, thereby realizing electromagnetic signal clustering based on high-order features of neural network signals. Figure 1 and Figure 2 As shown, the specific implementation process of the present invention is as follows:

[0028] Assume that for the nth (n=1,2,…,N) signal, the high-order features of the signal extracted by the neural network are represented as Feature n ; where N represents the total number of signals, Feature n The dimension of is K. For the mth (m=1,2,…,M) cluster, its clustering high-order features are expressed as Where M represents the total number of signals, Feature n and The dimensions of are all K-dimensional.

[0029] Initialization: The first signal is used as the first cluster, and its clustering high-order features are

[0030] Step 1: For the nth signal, assuming there are already m clustering results, calculate the cosine similarity between the nth signal and the mth clustering result in sequence, which is:

[0031]

[0032] where, represents the absolute value of the dot product between the high-order features of the nth signal and the mth clustering feature, ||Feature n || represents the Euclidean norm of the high-order features of the nth signal, represents the Euclidean norm of the high-order features of the mth clustering.

[0033] Step 2: Find R n,m The maximum value is MaxR n,m , and its corresponding serial number is max_m; set the threshold value to Thres. If MaxR n,m ≥Thres, then classify the nth signal into the max_mth clustering and proceed according to Step 3. If MaxR n,m <Thres, it is considered that the nth signal belongs to a new clustering and proceed according to Step 4.

[0034] Step 3: For the max_mth clustering, update its high-order clustering features with the high-order features of the nth signal, that is where α is the smoothing coefficient of the high-order clustering features. When the dot product between the high-order features of the nth signal and the mth clustering feature is greater than or equal to 0, otherwise

[0035] Step 4: Add a new clustering, that is m = m + 1, and set the high-order clustering features of this clustering to the high-order features of the nth signal, that is

[0036] Step 5: Starting from the 2nd signal, process according to Step 1 to Step 4 in sequence until all signals are processed, that is, complete the signal clustering processing based on the high-order features of the neural network signals.

[0037] In other embodiments of the present invention, assuming that for the nth (n = 1, 2,..., N) signal, the high-order features of the signal extracted by the neural network are represented as Feature n ; where N represents the total number of signals, and the dimension of Feature n is K-dimensional. For the mth (m = 1, 2,..., M) clustering, its high-order clustering features are represented as where M represents the total number of signals, Feature n and All dimensions are K-dimensional, and in this embodiment, K = 32.

[0038] Initialization: Take the first signal as the first cluster, and its cluster high-order feature is

[0039] Step 1: For the nth signal, assuming there are already m clustering results, calculate the cosine similarity between the nth signal and the mth clustering result in sequence, which is:

[0040]

[0041] where represents the absolute value of the dot product of the high-order feature of the nth signal and the high-order feature of the mth cluster, ||Feature n || represents the Euclidean norm of the high-order feature of the nth signal, represents the Euclidean norm of the high-order feature of the mth cluster.

[0042] Step 2: Find R n,m The maximum value is MaxR n,m , and its corresponding serial number is max_m; set the threshold value to Thres. If MaxR n,m ≥ Thres, then assign the nth signal to the max_mth cluster and proceed according to Step 3. If MaxR n,m < Thres, it is considered that the nth signal belongs to a new cluster and proceed according to Step 4. In this embodiment, Thres = 0.6.

[0043] Step 3: For the max_mth cluster, update its cluster high-order feature with the high-order feature of the nth signal, that is where α is the smoothing coefficient of the cluster high-order feature. In this embodiment, α = 1.

[0044] Step 4: Add a new cluster, that is, m = m + 1, and set the high-order feature of this cluster to the high-order feature of the nth signal, that is

[0045] Step 5: Starting from the second signal, process it according to Steps 1 to 4 in sequence until all signals are processed, that is, complete the signal clustering processing based on the high-order features of neural network signals.

[0046] The units involved in the embodiments described in the present invention can be implemented in software or in hardware. The described units can also be set in a processor. Among them, the names of these units do not constitute a limitation to the unit itself in some cases.

[0047] According to one aspect of an embodiment of the present invention, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in the above various optional implementations.

[0048] As another aspect, an embodiment of the present invention further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

Claims

1. A method for clustering electromagnetic signal streams based on high-order features of neural network signals, characterized in that: The following steps are involved: Suppose for the nth signal, the high-order feature of the signal extracted by the neural network feature is represented as Feature n , where n = 1, 2, ..., N, N represents the total number of signals, Feature n The dimension is K-dimensional; for the mth cluster, its clustering high-order features are expressed as Where m = 1, 2, ..., M, M represents the total number of signals, Feature n and The dimensions are all K-dimensional; the first signal is taken as the first cluster, and its clustering high-order features are After initialization is completed, follow these steps: S1: For the nth signal, assuming there are m clustering results, the cosine similarity between the nth signal and the mth clustering result is calculated in sequence, which is: in, Represents the absolute value of the dot product of the nth signal high-order feature and the mth cluster feature, ||Feature n || represents the Euclidean norm of the nth signal’s high-order features, Represents the Euclidean norm of the high-order features of the mth cluster; S2: Search for R n,m The maximum value is MaxR n,m , and its corresponding serial number is max_m; set the threshold value to Thres. If MaxR n,m ≥ Thres, then classify the nth signal into the max_mth cluster and proceed according to S3; if MaxR n,m < Thres, then consider the nth signal to belong to a new cluster and proceed according to S4; S3: For the max_mth cluster, update its cluster high-order features with the nth signal high-order features. Right now Where α is the clustering high-order feature smoothing coefficient. When the dot product of the nth signal high-order feature and the mth clustering feature is greater than or equal to 0, otherwise S4: Add a new cluster, that is, m=m+1, and let the high-order feature of the cluster be the high-order feature of the nth signal, that is S5: Starting from the second signal, process according to S1 to S4 until all signals are processed, that is, the signal clustering processing based on the high-order characteristics of the neural network signal is completed.

2. The electromagnetic signal stream clustering method based on high-order features of neural network signals according to claim 1 is characterized in that: The value of dimension K includes 32.

3. The electromagnetic signal stream clustering method based on high-order features of neural network signals according to claim 1 is characterized in that: The value of the threshold Thres includes 0.

6.

4. The electromagnetic signal stream clustering method based on high-order features of neural network signals according to claim 1 is characterized in that: The value of the clustering high-order feature smoothing coefficient α includes 1.

5. An electromagnetic signal stream clustering device based on high-order characteristics of neural network signals, 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 loaded by the processor, the method according to any one of claims 1 to 4 is executed.

6. A computer-readable storage medium, characterized in that: A computer program is stored in a readable storage medium, and the computer program is loaded by a processor to execute the method according to any one of claims 1 to 4.

7. An electromagnetic signal stream clustering system based on high-order features of neural network signals, characterized in that: Comprising the device as claimed in claim 5.

Images